gallego, 2006
Post on 10-Jul-2016
224 Views
Preview:
TRANSCRIPT
Voucher-School Competition, Incentives, andOutcomes: Evidence from Chile∗
Francisco A. GallegoDepartment of Economics, MIT
This Version: March 16, 2006First Version: February 1, 2004
Abstract
I investigate the effects of voucher-school competition on educational outcomes.In my model, parents decide between public schools and voucher schools (that is,private schools that receive a voucher for each enrolled student). The public schoolshave no direct incentive to produce quality, beyond meeting a minimum enroll-ment level and voucher schools face explicit competitive incentives. The modelproduces three empirical predictions: voucher-school competition 1) improves stu-dent outcomes; 2) may put stronger pressure on public schools to increase quality;and 3) has stronger effects when the minimum enrollment level is more binding.Since voucher school competition is endogenous to public school quality, I exploitthe interaction of the number of Catholic priests in 1950 and the institution of thevoucher system in 1981 as a potentially exogenous determinant of the supply ofvoucher schools. I document that the number of priests is determined historicallyand is correlated with the entry of both Catholic and non-Catholic voucher schoolsafter the voucher reform. Moreover, I document that the number of priests percapita has little effect on educational outcomes before the reform and a positiveeffect after the reform. Next, using priests in 1950 as an instrument for voucher-school entry in a cross-section of students in 2002, I confirm the main predictionsof my model: 1) an increase of one in the ratio of voucher-to-public schools in-creases tests scores by about 0.14 standard deviations; 2) the effects are similarfor students attending public and voucher schools; and 3) the effects are smallerfor public schools with less binding minimum enrollment levels.
∗Email address: fgallego@mit.edu. I would like to thank Daron Acemoglu and Dora Costa for theircontinuous advice and comments; Josh Angrist, David Autor, Miriam Bruhn, Ricardo Caballero, DanteContreras, Alexandre Debs, Amy Finkelstein, Julio Guzman, Jerry Hausman, Andres Hernando, DanielHojman, Caroline M. Hoxby, Chang-Tai Hsieh, Borja Larraın, Jin Li, Norman Loayza, Bruce Meyer,Arturo Ramırez-Verdugo, Casey Rothschild, Jose Tessada, Andrea Tokman, Sergio Urzua, BernarditaVial and seminar participants at the Central Bank of Chile, Dartmouth College, the Harris School ofthe U. of Chicago, MIT, Northwestern U., U.C.-Irvine, and Washington U.-St. Louis for comments; theMinistry of Education of Chile for access to data, especially Mauricio Jelvez and Claudia Matus; Sr.M. Jimena Alliende, Fr. Juan Dıaz s.j., and Br. Aldo Pasalacqua for information on Catholic schools;Gregory Elacqua for sharing information; and Donna Zerwitz for editing help. The usual disclaimerapplies.
1 Introduction
Creating competition between schools is a cornerstone of voucher school proposals. Pro-
ponents have argued that by creating competition, vouchers create stronger incentives
for public schools to increase quality. However, critics counter that school competition
may increase student segregation and harm poor students. In parallel, a line of research
on the effects of inter-school competition on student outcomes has not reached a con-
sensus on the causal effects on academic outcomes. While some papers find positive and
significant effects of school competition and school choice (e.g., Bayer and McMillan,
2005; Hoxby, 1994, 2000, 2005; Lavy, 2005; and Sandstrom and Bergstrom, 2005), other
papers do not find significant effects (e.g., Hsieh and Urquiola, 2004; Rothstein, 2004,
2005).
This paper contributes to the literature by studying the effects of inter-school com-
petition on the academic outcomes of Chilean students who attend publicly subsidized
schools. Chile is the only developing country that has operated the complete K-12 sector
under a “quasi-voucher” system for a long period of time (since 1981). Voucher schools
(that is, private schools that receive a voucher for each enrolled student) currently serve
above 40% of all students. However, voucher school enrollment varies widely across ar-
eas. While in 10% of the educational markets the voucher enrollment is more than 50%,
about 20% of Chilean municipalities have no voucher schools in operation.
This paper first presents a Hotelling-type model (Hotelling, 1929), in which parents
have heterogenous preferences for different schools, to analyze the effects on student
outcomes of having two types of schools in a market: public schools with no direct
incentives to produce quality beyond meeting a minimum enrollment level; and voucher
schools that face explicit competitive incentives. In this context, the model predicts
positive effects of voucher school entry on the quality offered by both voucher and public
schools (level effects). The model also predicts that the size of the response of public
schools to voucher school entry depends on the minimum enrollment level needed by a
public school to operate and on the size of the school age population (interaction effects).
Finally, the model suggests that voucher-school competition may put stronger pressure
to improve quality on public schools than on voucher schools.
As a potentially exogenous determinant of voucher school competition in different
markets, I exploit the interaction of the number of Catholic priests per person in 1950
in different areas of Chile with the establishment of the voucher system in 1981. I show
that the number of priests per person is historically determined; and, Catholic priests
are correlated with the entry of both Catholic and non-Catholic voucher schools after
1
1981. Enrollment in formally Catholic voucher schools increases from at most 5% of the
school-age population before the voucher reform to about 14% in 2002.1 In addition, I
present evidence that the entry of non-Catholic voucher schools is also related to priests
and present some hypotheses to explain this correlation. Moreover, before the 1981
reform, private school competition had no financial effects on public schools because
revenues of public schools did not depend in any way on enrollment in private schools.
Consistent with this, I also document that the number of priests per person had little
effect on educational outcomes prior to 1981 and has a positive effect on outcomes
after the voucher system was established. In other words, the potential validity of my
identification strategy relies on the assumption that Catholic priests were present in the
pre-voucher period, but that their effects on educational outcomes only became evident
during the period when the voucher system was established. The evidence supports this
view.
I estimate the level effects of the ratio of voucher-to-public schools on test scores in
an educational market for a cross-section of students in 2002. This sample of students
allows me to test the predictions of my theoretical model using data for the post-reform
period. Because my reduced-form results imply that priests had little effect on edu-
cational outcomes before 1981, I use the number of priests in different areas in 1950
as a potentially valid source of exogenous variation in the supply of voucher schools
during the post-reform period. I find that once I instrument for the ratio of voucher-
to-public schools in different educational markets, the entry of one voucher school per
public school in a market increases tests scores by about 0.14 standard deviations. The
results are similar for students attending both public and voucher schools, using differ-
ent measures of voucher school competition, and controlling for educational outcomes
before the voucher reform. In contrast, the OLS estimates do not show a strong effect
of voucher-school competition on test scores.
The data also support the existence of interaction effects of voucher school competi-
tion, as predicted by my theoretical model. Public schools located in municipalities that
have relatively low proxies for minimum enrollment levels and high education deficits do
not react strongly to voucher-school competition. These results constitute a more de-
manding test of the predictions of my model and are hard to be explained by alternative
theoretical models that may explain the level effects of inter-school competition. Some of
1Notice that by the Cannon Law of the Catholic Church, a school is formally recognized as Catholicwhen either (i) the Church directly appoints the school principal or (ii) the Church approves theapointment of the school principal. Therefore, many schools that are related to the Catholic Churchare not considered formally Catholic.
2
these alternative theories are as follows: (1) There may be direct effects of competition
on parents’ or other schools’ information; i.e., parents or schools use the information
provided by competitors to improve quality (Hoxby, 1994). (2) There may be reputa-
tion effects: yardstick competition among teachers who care about their performance in
comparison to other teachers (Juerges et al. 2004); this effect may be more relevant in
markets with more comparison points.(3) There may be poaching: good teachers sig-
nal their unknown characteristics and good (voucher) schools learn and hire these good
teachers. My results on the existence of the interaction effects of voucher-school compe-
tition are hard to reconcile with these alternative explanations and support the work of
my model stressing the role of incentives on the behavior of public school agents.
This is not the first paper studying the effect of voucher school competition in the
context of the Chilean voucher system. This paper contributes to this literature by
providing new IV estimates using a potentially valid source of exogenous variation for
voucher school entry (and providing a number of indirect tests to support its potential
validity) and presenting a theoretical model that allows me to study the mechanism
through which competition may affect the behavior of public schools. The contribution
in term of having potentially valid instruments is relevant because previously used in-
struments include variables like population size, urbanization rates, or socioeconomic
characteristics that likely do not meet the exclusion restrictions.
Previous papers include Carnoy and McEwan (1998) who use a panel of schools
from the early 1980s to the mid 1990s and find no effect of the share of voucher school
enrollment on average test scores at the school level. Given that the entry of voucher
schools may be endogenous, all the other papers in the literature present some instru-
mental variable approach to deal with the endogeneity problem. These papers could
be classified according to the time period studied. Hsieh and Urquiola (2004) use the
change in educational outcomes at the municipality level between 1982 and the early and
mid 1990s and find no effect of the change in voucher school enrollment on outcomes.
Even tough the focus of that paper is on OLS estimates, Hsieh and Urquiola (2004) also
present some IV estimates using the size of the population, urbanization rate, and the
inter-quartile range of education in 1982 as instruments for voucher school enrollment.
These results contrast with other papers analyzing the effect of school competition on
test scores: Gallego (2002) using school level data for 1994-1997 and IV estimates (the
instruments are urbanization rates and population size) finds a positive effect of voucher
school enrollment on test scores, Contreras and Macias (2002) find a positive effect of
competition (measured using a Herfindahl index) on pre-college tests for 1998 using
3
instrumental variables (population size, area, lagged Herfindahl index, and availability
of private schools, among others). Auguste and Valuenzuela (2003) is the paper most
directly comparable to Hiseh and Urquiola (2004). They present both individual and
market level regressions for the effect of voucher school enrollment on test scores using
data for the late 1990s and find a positive and significant effect of voucher school entry
on test scores. Differences in the time period studied are important to understand differ-
ences in results between the two papers. It is hard to argue that the system operating in
the 1980s was a real voucher system from the point of view of several agents: (i) public
school budgets were not affected by the voucher reform (in part because the decentral-
ization of public schools was not completed until the late 1980s), (ii) employment of
public school agents was quite rigid until the mid-1990s, (iii) test scores were not public
until the mid/late 1990s, (iv) local governments were not elected democratically until
1992, and (v) the real value of the voucher decreased steadily during the 1980s and only
recovered in the early 1990s. All these factors may explain why the effects are positive
in the 1990s (when the educational system was operating under rules that are closer to
the textbook version of a voucher system) and insignificant in the 1980s.2 My paper
presents additional evidence of positive effects of voucher school competition in the early
2000s.
The reminder of the paper is organized as follows. Section 2 briefly describes the
Chilean education sector. Section 3 presents a theoretical model for framing the em-
pirical analyses of the paper. Section 4 presents the data used in this paper. Section
5 describes the identification strategy. Section 6 presents the results of difference-in-
difference regressions using data on educational outcomes before and after 1981. Section
7 presents estimates of the level effects of voucher-school competition on student out-
comes using a cross-section of primary students in 2002. Section 8 presents estimates of
interaction effects and section 9 briefly concludes. An appendix presents the proofs of
the main theoretical results of the model.
2 Institutional Setting: Primary and Secondary Education in Chile
Before 1981, the government in Chile was involved in the funding and provision of edu-
cation, the supervision and regulation of curricula, the handling of human resources, and
2Another important difference between the 1990s and the 1980s is that test scores for the completepopulation are only available for the 1990s. The test scores available for the 1980s possibly cover ahighly selected sub-sample of schools in some areas (personal communication with Erika Himmel whowas in charge of the implementation of the early test scores). In addition, there are no pre-reform testscores available—notice that Aedo-Richmond (2000) argues that the reform started to be implementedabout 1978.
4
investment.3 The 1981 educational reform: transferred public education from the central
to the local governments (municipalities); established a per-student subsidy (voucher)
to be received by voucher and public schools depending on enrollment; allowed parents
to choose among any publicly-financed school; and allowed would-be schools to enter
the market.4
Three types of schools emerged: publicly owned schools (managed by local govern-
ments), voucher schools (owned by private agents), and non-voucher schools.5 The first
two types of schools receive vouchers; the non-voucher private schools do not receive
public funds, charge high tuitions, and serve upper- income students. In 2004, voucher
schools enrolled about 42% of students in 2002 up from an enrollment rate of 15% in
1981. Public school enrollment dropped from 78% in 1981 to 49% in 2002. The remain-
ing enrollment corresponds to non-voucher private schools, which I do not include in my
sample.
Public and voucher schools present important differences in terms of their incentive
structures and the amount of non-voucher resources they receive. Voucher schools tend
to behave like competitive firms, receiving revenues proportional to enrollment. While
some voucher schools are operated by for-profit firms, other voucher schools are operated
by non-profit organizations that raise additional funds in a relatively competitive mar-
ket for donations to be spent in schools (Aedo, 1998).6 In contrast, public schools work
under “softer” budget constraints: when needed, public schools that are losing students
receive transfers, above and beyond the vouchers to pay their expenses (Sapelli, 2003).
In addition, while vouchers were the only public intervention in the K-12 sector during
the 1980s, governments during the 1990s channeled additional resources to “vulnerable”
schools and increased non-voucher spending. My estimates are that about 30% of public
expenditure in education for the average student is not related to the voucher (data for
2002). In terms of teacher regulations, the public school teachers face flat wage sched-
ules and cannot be removed easily from their positions—only since the mid-1990s could
public school teachers be moved to other schools in the same municipality (Mizala and
3While the reform was formally enacted in 1981, Aedo-Richmond (2000) suggests that the reformbegan to be implemented de facto around 1978.
4Sapelli (2003) presents a more detailed description of the Chilean voucher system.5A small group of subsidized private schools did operate before the 1981 reform. These schools
enrolled about 7% of the school-age population (estimates using data from the 2002 Social ProtectionSurvey) and were financed with small public subsidies and private donations (Aedo-Richmond, 2000).Gregory Elacqua estimates that subsidized private schools enrolled about 12% of the students enrolledin schools in 1979 (personal communication). Considering that enrollment rate was about 75% in thesame period, subsidized private schools enrolled about 9% of the school-age population.
6Gregory Elacqua estimates that about 63% (58%) of voucher school students were enrollmed infor-profit schools in 1998 (1992). Notice that some for-profit schools are formally Catholic.
5
Romaguera, 2004). The number of teachers per student is 25% higher in public schools
than in voucher schools (CENDA, 2002). And, public schools are highly unionized.7 I
use these differences in the institutional environment that voucher and public schools
face in the model in Section 3.
The closing of public schools is a second institutional feature of the Chilean system
that I study in the model. The data suggest that about 8% of public schools stopped
reporting test scores during the 1990s. The most likely reason for this is that these
schools were closed or merged with other public schools. The closed schools tended to
have fewer students than other public schools and to under-perform relative to other
public schools. Interestingly, the teachers’ union is now actively lobbying against the
closing and merging of public schools. Moreover, the opinions of teachers reported in
Bellei et al. (2003) suggest that teachers do not want to be moved from one public school
to another. This evidence, in conjunction with the evidence on fixed wages, suggests
that public school teachers earn significant rents from working in public schools.
3 Motivating Theory
I present a simple theoretical model for studying the potential effects of voucher-school
competition on school quality. The model incorporates three groups of agents: parents
deciding among different schools, voucher school owners, and public school agents.
3.1 Agents
L parents are uniformly distributed over a linear neighborhood of length 1. The location
of parents along the linear neighborhood refers to their preferences for public and voucher
schools, which are located at the extremes of the city. In particular, let x denote location
along the linear neighborhood, public schools are located at x = 0 and voucher schools
at x = 1. This modelling approach formalizes the notion that public and voucher schools
not only offer formal instruction (i.e., measured in test scores), but also instruction in
other areas, such as religious education and civic values. I assume that parents have
heterogeneous preferences about these school characteristics and that schools cannot
choose their location along the linear city (i.e., public and voucher schools have intrinsic
differences in the non-formal instruction they provide).8
7The teachers’ union has strong political power and actively lobbies for additional benefits andagainst policies aimed at providing performance incentives.
8I could generate spatial differentiation in equilibrium by adding more structure into the cost func-tions (Tirole, 1988).
6
The utility of a parent located at x is given by:
Ujx = qj − tdjx, (1)
where qj is quality offered by the school j and djx is the distance from parents located at
x to school j. d captures the discrepancy between parents’ preferred instruction and the
instruction provided by school j. Therefore, t is a “transportation cost” that measures
the disutility per unit of distance of sending children to a school that is not located at
x. Parents choose the school that maximizes (1). If two voucher schools offer the same
quality, then parents randomize among them.
There is only one public school in each neighborhood, but the number of voucher
schools (N) is endogenously determined in the model, given some number of would-be
voucher schools in an area (NP ). The number of would-be voucher schools may be
limited, because school quality is a good whose reputation is important. School quality
is provided after students are enrolled; therefore, voucher schools may “hold up” parents
and not fulfill their initial offers of quality. In a context of incomplete contracts, only
voucher schools with good reputation or that can signal that they are not opportunistic
agents are able to enter the market and generate a positive demand. In addition, cultural
and social factors may constrain the number of acceptable voucher schools.9
Owners of voucher schools decide simultaneously whether they enter the market and
what quality they provide (qV ). Profits of voucher school i are given by:
Πi =£(v − cV (qVi))nVi − F
¤, (2)
where v is the per-student voucher, cV (·) is the unitary cost of providing quality qV , nVis the number of voucher school students, and F is a fixed cost. I assume cV 0 (·) > 0 andthat quality offered by voucher schools has to be above a minimum level qMV .
In terms of public schools, mayors (who manage public schools in a municipality) face
an agency problem; in contrast to voucher schools, public schools cannot use variable
wages and other forms of explicit incentives to implement their optimal choices. Thus,
public school agents (teachers) decide the quality that public schools provide (qP ) by
maximizing: ©y − cP (qP ) +R 1 [np ≥ n]
ª, (3)
9Survey data from Chile suggest that, indeed, there are limitations in the supply of voucher schools.Lehmann and Hinzpeter (1997) report that in 1996 about half of the parents with children in publicschools wanted to have their children in voucher schools. In 2001, while 39% of middle-income parentshad their children in voucher schools, 66% preferred a voucher to a pubic school. Results for low-income parents are similar: while 21% of these parents had their children in voucher schools, about50% preferred a voucher school.
7
where y is income of teachers, which is fixed, cP (·) is the effort cost of providing qualityqP (c
P 0 (·) > 0), 1 [nP ≥ n] is an indicator function that takes a value of one if nP ≥ n
and a value of zero otherwise, nP is the number of public school students, n is the
minimum enrollment level, and R is a positive constant that captures rents of public
school agents that are lost if the public school closes. The public school closes when its
enrollment is below the minimum enrollment level or its quality is below qMP .
The public school budget is:
[(v − o)nP − F +NV ] ,
where NV is a non-voucher transfer and o is a per student cost. I assume that only
non-voucher transfers vary across public schools. The minimum enrollment level of each
public school is the enrollment level associated with a balanced public school budget,
which is given by:
n =F −NV
v − o.
Since, in the model, only NV varies across public schools, n is determined by changes
in NV.10
3.2 Equilibrium
The timing of events is as follows:
1. First, a finite number NP of voucher schools simultaneously decide whether they
enter the market and the quality they provide (qV ).
2. Next, public school agents decide the quality they provide (qP ).
3. Finally, parents decide to which school to send their only child.
The symmetric subgame perfect equilibrium of this model is characterized by values
for qP , qVi , N, nP , nV , and nVi , such that:
• Quality offered by all voucher schools is the same, i.e., qVi = qV ∀ Vi.
• Parents maximize (1) by selecting among available schools.
• Public school agents maximize (3) by selecting public school quality.10Non-voucher transfers could be understood as soft budget constraints. In this case, the model in
Robinson and Torvik (2005) could be used to explain why in a political economy equilibrium, the centralgovernment can be willing to transfer public money to public school agents.
8
• N ≤ NP .
• Πi ≥ 0, and the N voucher schools that have entered the market maximize (2) by
selecting voucher school quality.
To solve the symmetric subgame perfect equilibrium of the model, I use backwards
induction, starting from parents’ choices. Parents choose a school to maximize (1).
Aggregating their decisions, I derive the demand for public and voucher schools, as
stated in Result 1 (proofs of the theoretical results are presented in the Appendix):
Result 1 Enrollment in the public school and voucher schools is, respectively:
nP =
½L if N = 0
L¡qP−qV2t
+ 12
¢if N > 0
,
nVi =
½0 if N = 0
LN
¡qV −qP2t
+ 12
¢if N > 0
∀ Vi.
School quality is determined by maximizing utility of public school agents, given
Result 1. Note that maximization of Equation (3) implies that either the minimum
quality or the minimum enrollment constraint is binding in equilibrium. Condition 1
states the case in which the minimum enrollment constraint binds.
Condition 1 (Minimum Enrollment Constraint Binding)
(n
L− 12) ≥ qMP − qV
2t.
Condition 1 is quite intuitive and states that the minimum enrollment level is more
likely to be binding when the minimum enrollment is high relative to the population
and the difference between minimum public school quality and voucher school quality is
large.
Result 2 presents the optimal public school quality.
Result 2 1. If Condition 1 holds, optimal public school quality is given by:
qP =
½qMP if N = 02t( n
L− 1
2) + qV otherwise
,
and, enrollment in public schools is given by:
nP =
½L if N = 0n otherwise
.
9
2. If Condition 1 does not hold, optimal public school quality is given by:
qP = qMP ,
and, enrollment in public schools is given by:
nP =
(L if N = 0
L³qMP −qV2t
+ 12
´if N > 0
.
The intuition of this result is that only one of the two constraints is binding in
equilibrium. Thus, if there is no voucher school in the market, then public school agents
offer the minimum quality, qPM . The same result emerges if the minimum enrollment
level is not binding. In contrast, if the minimum enrollment constraint binds, then
public school agents have to produce a (higher) quality level, such that the public school
has exactly the number of students needed to satisfy the minimum enrollment level.
Finally, regarding the behavior of voucher schools, I first determine the number of
voucher schools that have non-negative profits if they produce the minimum quality level
(qMV ), assuming no restriction on the number of would-be voucher schools. I denote this
number of schools as N∗. I analyze below the case in which there are a limited number
of would-be voucher schools.
Result 3 1. N∗ = 0 if nV (v − cV¡qMV¢) < F.
2. N∗ > 0 if nV (v − cV¡qMV¢) ≥ F. N∗ is given by:
N∗ :nV
(N∗ + 1)(v − cV
¡qMV¢) < F ≤ nV
N∗ (v − cV¡qMV¢).
This result highlights an important property of the model. Namely, that N∗ depends
positively on the total enrollment in voucher schools which, in turn, depends negatively
on public school quality (See Result 1). Therefore, this model predicts that the better
the public schools in an area are, the lower is the number of voucher schools operating
in that area. Even though the model is deterministic, when I move to the empirical
analyses, I could extend the model to have some randomness (e.g., in enrollments) and
get the result that there may be reverse causality from public school quality to the
number of voucher schools in an area. Therefore, OLS estimates of the effect of voucher-
school competition are downward biased. This bias suggests using instrumental variables
to identify the causal effect of the number of voucher schools on public school quality.
Next, the actual number of voucher schools depends on the availability of would-be
voucher schools in a market.
10
Result 4 The number of voucher schools operating in a market is:
N = min¡N∗, NP
¢.
Finally, the next result determines the optimal quality level offered by each voucher
school.
Result 5 If Condition 1 holds, quality offered by voucher schools is:
qV =
(qMV if N = 1 and NP = 1
cV−1³v − NF
L−n
´if N ≥ 1 and NP ≥ 2 .
If Condition 1 does not hold, quality offered by voucher schools is:
qV :
⎧⎨⎩cV (qV )
cV 0 (qV )= 2tnV
Lif N = 1 and NP = 1
qV = cV−1³v − NF
nV
´if N ≥ 1 and NP ≥ 2
.
This result highlights the role of competition in creating incentives for voucher schools
to increase quality. In the case when N = 1 and NP = 1, the single public school does
not face potential competitors, chooses the minimum quality level (qMV ) and earns rents
in equilibrium. In contrast, when NP ≥ 2 and N = 1, the existence of potential
entrants creates incentives for the incumbent voucher school to increase quality until
profits are equal to 0. Similarly, when N ≥ 2, the existence of potential competitorscreates incentives for incumbents to offer a quality level such that profits are equal to 0
in equilibrium.
Results 1 through 5 allow me to characterize the equilibrium quality given different
parameter values. This is presented in Proposition 1.
Proposition 1 (Equilibrium Public and Voucher School Quality) If Condition 1
holds in equilibrium, public and voucher school quality are given by:
qP =
⎧⎪⎨⎪⎩qMP if N = 0
2t( nL− 1
2) + qMV if N = 1 and NP = 1
2t( nL− 1
2) + cV
−1³v − NF
L−n
´if N ≥ 1 and NP ≥ 2
,
qV =
(qMV if N = 1 and NP = 1
cV−1³v − NF
L−n
´if N ≥ 1 and NP ≥ 2 .
11
If Condition 1 does not hold in equilibrium, public and voucher school quality are
given by:
qP = qMP ∀ N,
qV :
⎧⎨⎩cV (qV )
cV 0(qV )= 2tnV if N = 1 and NP = 1
qV = cV−1³v − NF
nV
´if N ≥ 1 and NP ≥ 2
.
To focus on the most interesting cases for the purposes of this paper and to reduce
the number of possible cases, I discuss mainly the case in which the minimum enrollment
constraint binds in equilibrium (Condition 1). This seems the most plausible case for
Chile.11 When Condition 1 is violated, public school quality does not depend on the
number of voucher schools. In turn, voucher school quality is an increasing function of
the number of voucher schools in the market. When Condition 1 holds, the minimum
enrollment level is binding; therefore, public school quality also depends on the number of
voucher schools in the market, because public schools respond to voucher school quality
in order to achieve the minimum enrollment level.
Proposition 1 allows me to study the effects of increases in NP on public and voucher
school quality. These comparative static exercises are closely related to my empirical
analysis in this paper, which uses a potentially valid source of exogenous variation of N .
In the remainder of the model, I only consider the case in which N ≤ 2, which is therelevant interval for Chile, because the ratio of voucher-to-public schools is less than 2
in 95% of the educational markets.
Corollary 1 (Level Effects) If Condition 1 holds in equilibrium and N ≤ 2, an in-crease in NP that increases N has a positive effect on equilibrium voucher and public
school quality .
If Condition 1 holds in equilibrium and N = 1, an increase in NP from 1 to 2 has a
positive effect on equilibrium voucher and public school quality.
Intuitively, when N increases from 0 to 1, the public school increases the quality
offered because the minimum enrollment constraint becomes binding. When N increases
from 1 to 2, the incumbent voucher school has to increase its quality until rents are
dissipated, because the entering voucher school offers a higher level of quality. In this
11Condition 1 is supported by the data. The Chilean data suggests that: −0.06 ≈ ( nL− 12) >
qMP −qV2t ≈
−.16−.031.61 ≈ −0.12. I estimate t from a regression of nP
Lon qP − qV . I approximate qMP using the average
test scores of public schools when N = 0, and qV using the average test scores for voucher schools inmarkets with N = 1.
12
case, the public school has to respond to the increase in voucher school quality in order
to meet the minimum enrollment constraint. The second case stated in Corollary 1
highlights the effects of increasing potential competition when the market is such that
only one voucher school operates. In this case, potential competition creates incentives
for the incumbent voucher school to increase quality until profits are dissipated.
Corollary 2 presents another interesting theoretical implication of an increase in N
that is caused by an increase in NP . This corollary resembles an empirically relevant
case for Chile because public schools in about 85% of the educational markets face
competition from at most one voucher school and, as previously mentioned, the ratio of
voucher to public schools is less than 2 in 95% of the educational markets. Thus, this
corollary produces theoretical predictions related to the variation I observe in the data.
Corollary 2 (Public vs. Voucher School Response to Competition) Let qNj de-
note equilibrium quality of school j when the number of voucher schools in a market is
N . Define zP and zV :
zP ≡ q1P − q0P , and
zV ≡ q2V − q1V
If Condition 1 holds in equilibrium and N increases initially from 0 to 1 and next
from 1 to 2 as a consequence of an increase in NP :
1. If zP > zV , public school quality responds more strongly than voucher school quality
to an increase in voucher school competition.
2. If zP = zV , public school quality and voucher school quality respond the same to
an increase in voucher school competition.
3. If zP < zV , public school quality respond less strongly than voucher school quality
to an increase in voucher school competition.
The comparative static exercise in Corollary 2 implies that voucher school competi-
tion could put stronger pressures to increase quality on public schools than on voucher
schools. Theoretically, there is no restriction to the relative response of both voucher
schools to the comparative static exercise presented in Corollary 2. For instance, public
school quality may be more responsive when the difference between minimum voucher
school quality (qMV ) and minimum public school quality (qMP ) is sufficiently large, which
produces a tremendous increase in public school quality in order to meet the minimum
13
enrollment level. On the contrary, if qMV − qMP is small or the minimum enrollment level
is sufficiently small, then voucher schools’ response to competition may be greater. The
empirical results in this paper suggest that both types of schools respond similarly to
voucher school competition.
Finally, Proposition 1 implicitly states that the response of the public school to an
increase in the number of voucher schools produced by an increase in NP depends on
how binding the minimum enrollment constraint is. The minimum enrollment constraint
is more binding when the school age population is large or when non-voucher transfers
are large. Corollary 3 states this result.
Corollary 3 (Interaction Effects) If Condition 1 holds in equilibrium and N in-
creases from 0 to 1 as a consequence of an increase in NP , then the lower the minimum
enrollment level (the more the non-voucher transfers there are), the lower is the response
of the public school to the entry of a voucher school in the market.
The predictions of Corollary 3 are quite intuitive. Public schools with more stringent
incentives in the form of harder enrollment levels have to increase quality more when
a voucher school enters the market. This corollary clearly illustrates the working of
the model in terms of the mechanism that makes voucher schools increase quality when
facing voucher school competition.
In summary, the most important empirical prediction of the theoretical framework
presented in this section is that public and voucher school quality should increase as the
number of voucher schools in a market increases exogenously. In addition, there should
be interaction effects: the public school response to exogenous changes in voucher school
competition depends on how binding the minimum enrollment is. This result illustrates
the theoretical mechanism at work in the model, which is basically related to the implicit
incentives that the minimum enrollment constraint gives to public school agents. In
addition, the theory predicts that the response of public schools to an increase in the
number of voucher schools in a market may be greater than the response of voucher
schools if the minimum public school quality is sufficiently low. Finally, the theory also
suggests that the number of vouchers in a market is endogenous to public school quality.
This suggests the use of instrumental variables in the empirical analyses of this paper.
4 Data
I use several datasets in this paper. Table 1 presents the variables used, the level at
which each variable is collected, and the descriptive statistics of each variable. I use
14
data on students’ educational outcomes, their backgrounds, parent preferences, school
characteristics, and the characteristics of the area where they attend school from the
dataset of the 2002 SIMCE (Sistema de Medicion de la Calidad de la Educacion) test,
which was administered to 4th graders. This test has been given nationwide since 1988
to more than 90% of students in a different grade each year (4th, 8th, or 10th graders).
I use the average of the Math and Spanish portions of the test (standardized to have an
average of 0 and a standard deviation of 1) as my measure of academic outcomes. I use
income per household member and mother’s education to measure the socioeconomic
background of students.12
Second, I use the CASEN (Encuesta de Caracterizacion Socioeconomica) 2000 survey,
which collects information on socioeconomic variables for a representative cross-section
of the population. I use a high school graduation dummy as a measure of educational
attainment for members of different cohorts that attended school in different places.
Third, I use the 2002 Social Protection Survey (called “Labor History and Social Secu-
rity”), which collects life-time information for a sample of individuals. I use information
on high school graduation rates at the market level for individuals attending school
before the 1981 reform, the migration decisions of parents with school-age children in
2002, and information on the type of school attended (public, subsidized private, and
paid private).
I measure the degree of voucher school competition as the ratio of voucher schools to
public schools in each educational market. I use 297 municipalities and the Metropolitan
Area of Santiago as proxies for local educational markets. Municipalities are separate
educational markets because, with the exception of municipalities in the Metropolitan
Area of Santiago, most students attend schools in the town where they live (Hsieh and
Urquiola, 2004; and Sapelli and Vial, 2002). Data on the availability of schools in each
market come from the Ministry of Education files.
Data on religious variables at the diocese level related to my identification strategy
come from the yearly publication by the Vatican called Annuario Pontificio (the number
of priests, the share of Catholics, and the ratio of order to total priests in each Chilean
diocese).13
12I use five categories to measure mother’s education (having attained at most primary education,secondary general education, secondary technical education, post-secondary technical education, andcollege or postgraduate education).13In particular, I use data on priests per-capita for 1950 from Annuario Pontificio. I estimate the
number of priests in the different dioceses in 1950 by considering the territorial division existing in the1990s (which includes 26 dioceses) and the number of priests in 1950. I make this adjustment becausesome dioceses (namely Santiago) included disproportionately big areas of population in 1950. Betweenthe 1960s and the 1990s, new dioceses appeared when some dioceses were split up–so the number of
15
I also use other sources of data in some empirical exercises. Data on Catholic schools
come from the school directory of the Chile Catholic Church (http://www.feducech.cl/
and http://www.iglesia.cl/). Data on municipal variables such as expenditures per stu-
dent and the size of public schools come from the Chilean Municipal Dataset (available
at http://www.sinim.cl/). Finally, I use information on electoral outcomes at the mu-
nicipality level from the Chilean Electoral Office, when analyzing the interaction effects
of inter-school competition.
5 Identification Strategy
One major challenge for an empirical analysis of the relationship between voucher school
competition and educational outcomes is the potential endogeneity of the number of
voucher schools. In this section, I argue that the interaction of the number of Catholic
priests per person in 1950 and the school reform of 1981 allows me to identify the
exogenous variation in the number of voucher schools in different educational markets.
The potential endogeneity between the number of voucher schools and public school
quality can be illustrated using the model in Section 3, in a context with randomness.
Increases in minimum quality (qMP ) or minimum enrollment (n) decrease the number of
voucher schools operating in an area and have a positive effect on quality offered by public
schools. Thus, entry into the market is endogenous to public school quality; therefore,
simple correlations or OLS estimates will produce downward biased estimates of the
causal effect of voucher-school competition on educational outcomes. Alternatively, OLS
estimates could be biased upward if voucher school entry responds to some unobserved
(to the econometrician) characteristic of the market that has a positive effect on school
quality.
My identification strategy exploits the interaction of the (log of the) number of
Catholic priests per person in 1950 and the 1981 reform to identify the exogenous vari-
ation in the number of voucher schools in an area, after controlling for the share of
Catholic population.14 The basic motivation for this identification strategy is two-fold.
On the one hand, there are direct effects of priests on Catholic schools. The involve-
dioceses increased from 19 to 26. I assume that the distribution of priests within the split dioceses isgiven by the distribution when the new dioceses were created (the distribution of priests within diocesesin the following periods is quite stable). In all the empirical applications, I cluster standard errors atthe 1950 dioceses level.14By controlling for the share of the Catholic population, I take into account potential direct effects of
this variable on educational results (as suggested by recent research on the effects of religious affiliationon income, education, and other social and economic variables, e.g., Barro and McCleary, 2003 andGruber, 2005).
16
ment of priests in educational activities is understood as a key element of their religious
mission (Garrone, 1977). However, although Catholic priests were involved in schools
before the reform (my estimates are that at most 5% of the school-age population at-
tended Catholic schools that were publicly subsidized),15 there is a big increase in the
enrollment in Catholic voucher schools after the reform, increasing to a level close to
14% of the school-age population in 2002. This increase takes place along three margins:
(i) new Catholic schools were installed, (ii) existing Catholic schools increased in enroll-
ment, and (iii) former paid Catholic schools became voucher schools after the reform.16
17 The main reason for this big increase in enrollment in Catholic voucher schools is
that only after the reform did Catholic schools start to receive vouchers (the voucher
represented an increase of about 160% in the value of the subsidy received by Catholic
schools from the government),18 allowing priests to establish new schools or to expand
enrollment among middle- and lower-class students.
Priests are important actors in Catholic schools. These schools can be owned directly
by the Church, by religious orders, or by people supported by the Church, but they
always have at least one priest acting as a chaplain. Currently, priests tend to focus on
pastoral ministries (e.g., being chaplains and teaching religious education classes) and on
the management of schools. In terms of time requirements of being a priest in a school,
priests have to spend a significant amount of time working with students, teachers, and
parents. Priests working in schools receive wages that are comparable to those of other
teachers. Pasalacqua (2004) reports that about 5% of the teaching staff and 10% of the
15These estimates are computed as follows: the Social Protection Survey implies that 7% of themembers of pre-voucher cohorts attended subsidized private schools. Espinola (1993, quoted in Hsiehand Urquiola, 2004) reports that 53% of pre-reform subsidized private schools were Catholic. If Catholicand non-Catholic schools have the same size, then I estimate that 3.5% of the school-age populationattended Catholic subsidized schools before the reform. However, I know that today Catholic schoolstend to be bigger than non-Catholic schools. Then, using the ratio of size between Catholic and non-Catholic schools in 2002, I estimate that about 4.7% of the school-age population before the reform wasin Catholic subsidized schools.16Unfortunately, there are no systematic data on enrollment in Catholic schools before and after the
reform. My estimates, using data from Pasalacqua (2004), are that 25% of enrollment in Catholicvoucher schools in 2002 corresponds to brand new Catholic schools and between 15 and 25% of theenrollment corresponds to schools that became voucher schools after the reform.17Anecdotal evidence from one of the most important Catholic groups in Chile (the Marist brothers)
may help to understand the increase in enrollment in Catholic voucher schools after the reform. En-rollment in Marist schools was 5,000 students in 1980, with about 10% of these students in subsidizedschools. In contrast, enrollment was 14,800 in 2002, with 40% corresponding to voucher schools. Abouttwo thirds of enrollment in voucher schools corresponds to schools established after the reform (Personalcommunication with Br. Aldo Pasalcqua).18This value is computed as follows. The intial value of the voucher was 30% higher than expenditure
per student in public schools before the reform. Before 1981, private schools received on average 50%of public schools expenditure per-student (Hsieh and Urquiola, 2004). Therefore, the nominal value ofthe voucher increased by 160%.
17
non-teaching staff in Catholic voucher schools are religious personnel (including not only
priests, but also brothers and nuns).
In addition, there may be effects of priests on the establishment of non-Catholic
schools. First, as previously discussed, the formal definition of a Catholic school is
restrictive; therefore, there may be some schools that are somewhat informally related
to the Church.19 Second, some non-Catholic voucher schools may have been established
by former teachers of Catholic schools (or even priests/nuns that retired from their
religious orders). A slight variation of this mechanism is non-Catholic voucher schools
that establish religious practices, names and curricula in their schools to try to mimic
Catholic schools in the same area. And, third, the propensity of parents to send their
children to private schools may be affected by the presence of old Catholic schools (and,
therefore, priests) in the same area. I do not have systematic evidence of these three
channels, but I present evidence in the next two sections that (i) priests are correlated
with the entry of non-Catholic schools, and (ii) priests do not affect the propensity to
attend Catholic vis-a-vis non-Catholic voucher schools at the individual level.
In addition, there is a fundamental change in the incentive scheme related to the
1981 reform. Even if enrollment in Catholic schools were constant after the reform, the
effect of priests on the behavior of public schools changes dramatically: from a system
where public school funds did not depend at all on enrollment in private schools to
a system where total funds are fixed and are allocated proportionally to enrollment.
The actual degree of school choice should have changed dramatically during the reform
period even if the enrollment in private schools were the same. This further motivates
using the interaction of priests in 1950 and the establishment of the voucher system as
an instrument for voucher school competition and expecting the effects of priests to be
significantly bigger over the voucher period.
Finally, the number of priests per person is historically determined and varies widely
across Chilean dioceses, despite the majority of the population being Catholic.20 While
the average diocese has a ratio of about 0.15 priests per 1,000 people, the diocese with
the highest ratio (0.23 priests per 1,000 people) has more priests per person than most
19Again, I have no systematic data to quantify this hypothesis, but two examples may help under-standing the magnitude of this phenomenon. About 40% of enrollment in voucher schools related tothe Jesuits (through Red Educacional Ignaciana and Fe y Alegria Chile) is not counted as enrollmentin Catholic schools. Most of these schools were established after the reform. Similarly, enrollment inthe (two big) Opus Dei voucher schools in Santiago is not counted as in Catholic schools because OpusDei schools are not related to the local bishop.20In the Catholic Church, a diocese is an administrative territorial unit, composed of many parishes
and governed by a bishop. Technically, each diocese is independent of the others and the bishop onlyresponds to the Pope.
18
Latin American countries; and, the diocese with the lowest ratio (0.06 priests per 1,000
people) is comparable to what is observed in a poor (and non-Catholic) country such as
Kenya. The variation in the number of priests across Chilean dioceses has to do mainly
with the fact that religious orders are more numerous in some areas than others.
Religious orders established themselves in a non-uniform way in different dioceses
in the past. The allocation of orders to different dioceses was related mainly to the
desire to bring priests to some Chilean areas beginning in the mid nineteenth to the mid
twentieth centuries (Aliaga, 1989; Araneda, 1986; Barrios, 1992). Some areas ended up
with more order priests because the bishops of some dioceses either belonged to orders
themselves or were more open to receiving order priests. While 71% of the order priests
in Chile in 2002 belonged to orders that entered the country between 1810 (the year of
independence) and the 1950s, only 5% of these priests belonged to orders that entered
the country after the 1950s.21 In general, there are more order priests than non-order
priests, conditional on the establishment of order in a place. Therefore, dioceses where
religious orders work tend to have more priests. The historical roots of the presence
of orders in different areas creates a positive correlation between the number of priests
today and in the past: the correlation between the number of priests per capita during
the 1990s and the 1950s is 0.78.
Table 2 illustrates the importance of orders in explaining the cross-diocese variation
in (the log of) priests per capita in 1950. I use the ratio of order priests to the total
number of priests as a proxy for the presence of orders in different dioceses. Results in
columns 3 and 4 support the claim that the variation of priests per capita is related to
the presence of orders in different dioceses.
I use priests per capita in 1950 as the main instrument for voucher school entry in
different areas during the voucher period, and the ratio of order-to-total priests as an
alternative instrument in some regressions. The next section empirically studies the
validity of this identification strategy.
6 Reduced Form Estimates: Difference-in-Difference Analysis andRobustness Checks
To validate the identification strategy discussed in the previous section, I need to show
that priests are not related to educational outcomes before the voucher reform and are
related to educational outcomes after the reform. Table 3 presents the basic evidence
supporting this identification strategy. In particular, I focus on three different cohorts.
21These numbers were computed using information from the 2002 directory of the Catholic Churchin Chile.
19
People older than 37 in 2000 (the year the CASEN survey was collected) attended school
before the reform was implemented; people between 26 and 37 years attended school
between 11 years and one year after the reform was implemented; and those less than
26 received their complete K-12 education after the reform was implemented. Because
I do not have data on test scores before the reform, I use high school graduation as a
proxy for school quality.
The results in Table 3 suggest that the number of priests has no relationship to high
school graduation rates for people attending school before 1981 (column 1). For the
second group (i.e. those receiving only a share of their primary and secondary education
after 1981), the effect of priests is slightly bigger but still not significant. Finally, for
those receiving their entire education after 1981, priests have a positive and marginally
significant effect on educational outcomes. Overall, this evidence supports the idea that
priests are correlated with educational outcomes only for the cohorts that attended
school after the 1981 reform.
Figure 1 presents a more detailed exercise to evaluate the differentiated effects of
priests on educational outcomes for cohorts that have different degrees of exposure to
the 1981 reform. This figure plots the relation between high school graduation and
priests in the decade closer to school attendance for individuals of different ages in
2000. This exercise also allows me to study whether the effect of the reform varies for
individuals of different ages. The results suggest that priests are only correlated with the
level of education for cohorts that attended school after the reform was implemented. As
importantly, the effect of priests on high school attainment is increasing in the number
of years that people attended school after the reform.
Overall, these results demonstrate the claim that priests only are correlated with
educational outcomes after the 1981 reform, and they confirm the rationale presented
in Section 5. Having established this central result, I present a number of additional
exercises to validate my identification strategy. Columns (1) and (2) in Table 4 presents
estimates of the relationship between priests per capita and the ratio of order-to-total
priests in 1950 and the ratio of voucher-to-public schools in each market in the voucher
period. Priests and the ratio of order-to-total priests both have a positive and signifi-
cant effect on the number of voucher schools per public school. The effect of Catholic
affiliation is also positive and significant, as expected.22 Column (3) presents the same
regression but using voucher school enrollment as the measure of voucher school entry
22Results for other variables included in these regressions are similar to other papers in the literature:mean (standard deviation of) education and income have positive (negative) effects on the availabilityof voucher schools, and more populated areas have more voucher schools.
20
(as in all the other papers in the literature). Again, priests have a positive and significant
effect on voucher school enrollment.
Column (4) presents estimates of the effect of priests on the change in public school
enrollment from the pre-reform period. I was able to compute public enrollment rate for
52 geographical areas for 1975 from MINEDUC (1975)—the geographical classification
corresponds to the split between urban and rural of 26 provinces covering the complete
Chilean territory. I define the change in enrollment in public schools as enrollment
today minus enrollment in 1975 in the area where the school is located. Results imply
that public school enrollment decreased more in areas with more priests, confirming my
argument in Section 5. The last three columns study the correlation between priests and
Catholic and non-Catholic voucher schools. The results imply that priests are correlated
with both the entry of Catholic and non-Catholic voucher schools. Moreover, the ratio
of Catholic to non-Catholic schools does not seem to be affected by priests. Thus,
these results confirm my argument of the potential effects of priests on the entry of
non-Catholic voucher schools.
I present additional evidence of the differential effect of priests after and before the
voucher reform in Table 5. This table studies the effect of priests on the attendance
to voucher schools for members of cohorts that attended school after and before the
voucher reform. Data come from the 2002 Social Protection Survey. I study whether
priests have a different effect on the likelihood of attending a voucher school before
and after the reform, controlling for a dummy for urban areas, and region and age
dummies. Results both coming from linear probability and probit models suggest that
priests are more related with the decision to attend a voucher school after the reform, as
expected. Results for the effect of priests on private school attendance before 1981 are
inconclusive: the linear probability model suggests a positive and significant effect, but
the probit model suggests an insignificant effect. Overall, evidence in this table confirms
the hypothesis that priests are more important to explain voucher school attendance
over the voucher period.
One potential concern for my identification strategy is that priests may have af-
fected education outcomes after the reform through other channels than voucher-school
competition. In this context, I study whether immigration decisions of families with
school-age children are correlated with the presence of priests in different areas. If in-
migration rates of families with children are higher in areas with more priests, then an
alternative explanation may be sorting of families based on taste/motivation for educa-
tion. I do not expect this channel to be important, given the available evidence showing
21
that migration in Chile is low because of public housing policies (Soto and Torche, 2004;
Tironi, 2003). The results in Table 6 confirm this presumption and show that both
micro estimates of in-migration decisions of families with children and macro estimates
of immigration rates at the region level do not support the view that in-migration rates
are higher in areas with more priests.
Overall, I document that the number of priests per person is historically determined;
prior to 1981, it has little effect on educational outcomes. I also show that Catholic
priests affected educational outcomes after the voucher system was established and are
correlated with several proxies of the levels of and changes in voucher school entry. These
results suggest that I have a potentially valid source of exogenous variation in the supply
of voucher schools in different areas during the voucher period. Using these results, the
next section estimates the level effects of voucher-school competition on student test
scores in a cross-section of students in 2002.
7 Estimating Level Effects
In this section, I present the results of regressions using information on test scores from
a cross-section of students in the voucher period. This approach has several advantages
over the reduced form estimates I presented before. First, I have detailed information on
the degree of voucher-school competition in the educational market where the student
attends school. Second, I have a more direct measure of school outcomes (test scores)
than in the previous exercises (high school graduation), which allows me to estimate
more precisely the effect of voucher-school competition on test scores. Third, I am able
to study whether the interaction effects predicted by the model are supported by the
data. Since the results in Section 6 suggest that priests only affect educational outcomes
after the reform, the number of priests per capita prior to 1981 is a valid instrument for
voucher-school competition during the reform (as well as for the interaction of priests
with the reform). Thus, I estimate the impact of voucher-school competition on student
test scores by running a regression of the form:
qim = πRm +X0
imα+M0
mβ + Y0
mρ+ εim. (4)
Subscript i refers to students and m to educational markets. q is test scores, R is the
ratio of voucher to public schools in market m, X is a vector including pre-school char-
acteristics of students (mother’s education and log of income per household member),
M is a vector including the mean and standard deviation of mother’s education and
income at the market level, Y is a vector including exogenous variables (Catholic popu-
22
lation, total school age population, urbanization rate, and region dummies), and ε is a
student-specific error term.23
Following the model presented in Section 3 , I use the ratio of voucher-to-public
schools as my measure of voucher-school competition–Rm in equation (4)–at the mar-
ket level. When I compute the number of voucher schools per public school at the
market level, I obtain the average availability of voucher schools per public school in
each neighborhood (assuming that one public school exists in each neighborhood, which
is reasonable in the case of Chile).
I estimate equation (4) using the log of Catholic priests per person in 1950, or the ratio
of order-to-total priests (only in some regressions to save space), as my instrument for
Rm. In addition, as discussed in Section 5, the ratio of order-to-total priests corresponds
to a more basic source of variation in the number of priests in different areas in Chile.
Thus, I also present IV estimates using the ratio of order-to-total priests as an alternative
instrument (in one specification I use both variables as instruments for voucher-school
competition). I use variables measured in 1950, which corresponds to the end of the
period of entry of Catholic orders into Chile.
7.1 Test Score Regressions
I first estimate equation (4) using the complete sample of students attending public and
voucher schools in the 2002 SIMCE dataset. Table 7 presents OLS estimates (columns
1-2) and IV estimates of equation (4) using log of priests in 1950 (columns 3-4) and
the ratio of order-to-total priests (columns 5-6) as my instruments for Rm. In each
case, I present first a parsimonious representation of the regression without including
controls and only including my measure of voucher-school competition. Next, I include
market- and student-level controls. The IV estimates are larger than the OLS estimates
as suggested by my model, because school entry may be endogenous to quality. While
the OLS estimates are -0.02 and 0.04, the IV estimates are in the interval between 0.13
and 0.17.24 The IV point estimates are always positive and significant. These results
imply that the effects of voucher-school competition on test scores are also economically
relevant. An increase of one voucher school per public school in a market is associated
with an increase in test scores of between 0.13 and 0.17 standard deviations. This is
23Since I include multiple observations of variables in the same area, I use the White/Huber estimatorof the variance-covariance matrix to compute corrected standard errors that are robust to arbitraryheteroskedasticity and clustered standard errors.24Measurement error in voucher-school competition may also explain why my OLS estimates are
smaller than my IV estimates.
23
equivalent to about half of the effect of increasing mother’s attainment from primary to
secondary general education.
Estimates for other variables included in the regression have the expected signs. All
socioeconomic controls are significant and have the expected sign: students with more
educated mothers tend to perform significantly better, and students from households
with higher incomes have higher test scores. The only two market-level variables with
a statistically significant effect are mean per capita income (a positive effect) and the
school-age population of the market (negative effect). The effect of the share of Catholics
is negative, but it is not precisely estimated.
Next, column (7) in Table 7 presents estimates using both priests and the ratio of
order-to-total priests as instruments. As expected, the estimated effect falls between the
estimates in columns (4) and (6). More importantly, an over-identification test of this
specification does not reject the null hypothesis that the instruments are valid. Formally,
the result of the over-identification test implies that IV estimates using each of the two
instruments separately are not statistically different among them.
These estimates of the effect of voucher school competition on test scores are in the
range of the estimated results of previous studies for Chile. My estimates imply that
a one-standard deviation increase in voucher school competition increases test scores
by about 0.10 standard deviations. Auguste and Valenzuela (2003) find an effect of
the same size at the student level, Contreras and Macias (2002) report an effect of a
one-standard deviation increase in the Herfindahl index in the interval of 0.08 and 0.17
standard deviations of test scores, and Gallego (2002) presents estimates that imply
an increase of between 0.03 and 0.18 standard deviations of the test scores when his
measure of competition increases by one-standard deviation. In contrast, my estimates
are slightly smaller than the effects of inter-school competition reported in papers for the
US. Bayer and McMillan (2005) and Hoxby (2000) report that a one-standard deviation
increase in their proxies for the degree of inter-school competition increases test scores
by about 0.15 standard deviations.
The estimates on the effect of voucher-school competition on test scores also are con-
sistent with my reduced-form estimates of the effects of priests on high school graduation
rates reported in Section 6. My reduced-form estimates in the previous section imply
that a one-standard deviation increase in the (log of the) number of priests (roughly
equivalent to 43 log points) increases high school graduation by about 11 percentage
points for people attending school after the reform (computed using results from Figure
1). In turn, a similar increase in the number of priests increases the ratio of voucher-
24
to-public schools by about 0.22 (Table 4), which increases test scores by between 0.031
and 0.037 standard deviations. To relate both results, I use the estimates of Hanushek
and Kimko (2000), who find that a one-year increase in schooling is associated with
an increase in about 0.042 standard deviations in test scores. In my sample, a person
who graduated from high school has about six years more of education than the rest
of the population. Putting everything together, I expect the impact of the increase in
high school graduation to be consistent with an increase in test scores of 0.03 standard
deviations (0.11×0.042×6 ≈ 0.028). Therefore, reduced-formed estimates of the impactof priests on high school graduation and cross-section estimates of the effects of voucher
school competition on test scores are consistent.
Table 7 presents additional specifications in which I replace my measure of voucher-
school competition by three alternative measures: the share of enrollment in voucher
schools (Auguste and Valenzuela, 2003; Gallego, 2002; Hsieh and Urquiola, 2004), the
Herfindahl index (Contreras and Macias, 2002), and the change in enrollment in public
schools between the pre-reform (1975) and post-reform periods (2002). Results in the
last three columns of Table 7 imply a positive effect of the three alternative measures
of voucher school competition on test scores. The size of the effects of a one-standard
deviation increase of competition on test scores varies across specifications: while the
implicit estimates in columns (8) and (9) imply an increase of test scores of about 0.15
standard deviations, estimates in column (9) imply a probably unreasonable increase of
test scores of 0.48 standard deviations.
I study several specifications checks in Table 8. First, I study one potential concern
with the results in Table 7: some variables, such as income and mother’s education, may
be affected by my instruments if parents attended school after the reform. If mother’s
education and income do not involve measurement error, then my earlier estimates can be
interpreted as the direct effects of voucher-competition on student outcomes. However,
if mother’s education and income are subject to measurement error, then the estimates
of voucher-school competition may be biased upward.25 To deal with this potential
problem, Column (1) presents the results excluding students with parents who attended
school after the reform (i.e. parents older than 39 years). The results are very similar
to those shown in Table 7, suggesting that my previous estimates do capture the direct
effect of voucher-school competition on test scores, controlling for mother’s education
and income.
Next, Column (2) presents the results of excluding schools located in rural areas.
25More precisely, mother’s education and income may not capture all the effects of mother’s humancapital on children’s educational outcomes.
25
The implicit assumption in the model in Section 3 applies for students that have low
physical transportation costs of moving from one school to another. I expect the effects
of competition to be significantly smaller if students cannot move easily from one school
to another, as expected in rural areas. Results in Column (2) confirm this idea: the point
estimate of inter-school competition increases when excluding rural schools. The next
two columns deal with the potential effect on the results of the Santiago Metropolitan
Area. Santiago is at the same time the biggest market (includes about 28% of the
students in my sample) and one of the markets with the highest ratio of voucher to
public schools (2.41 voucher schools per public school). To deal with the size effect
without losing information on a market that has a high degree of choice, I report in
columns (2) and (3) the results of running 50 regressions excluding randomly 90% and
50% of the Santiago students. Point estimates are smaller but positive, statistically
significant, and in the confidence interval of the estimates for the complete sample.
Next, Columns (5) to (7) present the results of an additional exercise: I include con-
trols for systematic differences in pre-reform educational outcomes in equation (4). I use
three proxies for pre-reform outcomes: (i) high school graduation rate in public schools
for cohorts that attended school before 1980 from the Social Protection Survey (column
5), (ii) high school graduation rate at the municipality level for cohorts that attended
school before 1980 from the CASEN survey (column 6), and (iii) average 1991 SIMCE
test scores at the municipality level (column 7), (indeed, this test score was applied 10
years after the reform was implemented, but at the same time (i) 1991 corresponds to
the first year for which I have complete information for all the schools after the decen-
tralization of the management of public schools was completed and (ii) as previously
discussed, it is hard to argue that a real voucher system was in operation until at least
the mid 1990s). Even though my results in Section 6 suggest that my instruments are
not correlated with pre-reform differences in academic outcomes, these results provide
an additional check. The results show that my main estimates of the effect of the ratio
of voucher-to-public schools change slightly in value, but remain statistically significant.
The point estimates for the effect of voucher school competition decrease slightly in
columns (6) and (7), and increase slightly in column (5).
Next, column (8) of Table 8 introduces controls for the composition of students at the
school level: mean and standard deviation of mother’s education and per-capita income.
I follow Hoxby (2000) in including these variables without giving a formal interpretation
to the estimates I find, and, therefore, I only focus on the effect of including these
variables on the estimate of voucher-school competition on test scores. Results imply
26
that the point estimate of the effect of voucher school competition is basically unchanged
with respect to most other estimates I present in Tables 7 and 8. Unfortunately, I do
not have additional valid instruments to study the causal effects of peer-effects on test
scores, but results in this column are at least suggestive that peer-effects are not driving
my main results.
I take an alternative way of analyzing whether peer-effects are driving my results in
the last two columns of Table 8, where I present estimates for sub-samples of students
in public and voucher schools. These sub-samples allow me to analyze whether the
estimated effect of voucher-school competition on student outcomes in public schools is
different than for voucher schools. These exercises have interest from two perspectives:
(i) I am able to study empirically the predictions of Corollary 2 and (ii) I am able to
test the conjecture of some papers that voucher-school competition may increase average
test scores but students attending public schools may be harmed if voucher schools are
able to cream-skim. Indeed, I cannot disentangle the contributions of both perspectives
without an additional source of exogenous variation.
Since I estimate (4) for sub-samples of students, I implement a Heckman selection
model with endogenous variables (Wooldridge, 2002). This procedure allows me to con-
trol for potential selection bias if the students included in these regressions are not
randomly selected from the population. To implement this procedure, I need to find
a variable that affects the selection of students in different schools and has no direct
effect on test scores. My instrument in the selection equation is a dummy that takes a
value of one if the teaching of values was among the top three criteria used by parents
for choosing schools. Since the mention of ”teaching of values” (i.e., la ensenanza de
valores in Spanish) has a religious connotation in Chile, this variable may capture rel-
ative preferences for voucher vis-a-vis public schools, or Catholic vis-a-vis non-Catholic
voucher schools.26 In terms of my model, the values variable is a proxy for the location
of parents in the linear city.
In the initial stage of the estimation, I run a selection equation of the following form,
using probit:
pim = 1³ϕSim + ζZm +X
0imξ +M
0mκ + Y
0mτ + µim > 0
´, (5)
where pim takes a value of 1 if an observation is included in the regression of interest, Sim
is some variable that affects pim and is not included in equation (4), Zm is an instrument
26Other papers have found that parents’ preferences for the teaching of values affect the choice ofprivate versus public schools in the US (Sander, 2001). In a related result, Howell (2004) finds thatreligious identity affect the participation in a voucher program in New York.
27
for Rm (i.e., priests per capita), and µim ∼ N(0, 1). Notice that in this equation I include
all the right-hand side variables included in the first stage of (4). Next, for the selected
subsample (i.e., for observations with pim = 1), I estimate the equation:
qim = πRm +X0imα+M
0mβ + Y
0mρ+ ψλim + εim
by 2SLS, excluding Sim from the second stage regression. λim is the estimated inverse
Mill’s ratio for each observation.
Table 9 presents probit estimates of the marginal effects of the variables included in
equation (5) on the choice between attending a public versus a voucher school and the
choice between a non-Catholic versus a Catholic voucher school. The results indicate
that the values variable is significantly correlated with the decision to send the student
to a non-Catholic school. Marginal effects imply that if parents care about the teaching
of values, they are 36% less likely to send their children to a public school versus a
voucher school, and are 28% less likely to send their children to a non-Catholic voucher
school vis-a-vis a Catholic voucher school.
Estimates for other variables included in equation (5) confirm previous results in the
literature on the socioeconomic determinants of attending a public school (e.g. Sapelli
and Vial, 2002 and 2003 for Chile; Checchi and Jappelli, 2004 for Italy): mother’s ed-
ucation and family income have a negative and significant impact on the probability of
attending a public school. Other estimates in column (1) suggest that market character-
istics are also important: the probability of attending a public school drops in urban and
more populated areas, in poorer areas, and in areas where Catholic affiliation increases.
Results for the probability of attending a non-Catholic versus a Catholic voucher school
indicate that education and income play a similar role to the choice between a public
and a voucher school, although the estimated effects are smaller and less significant.
Similarly, the probability of attending a non-Catholic voucher school decreases in areas
with a larger Catholic population and increases in more populated markets. Interest-
ingly, as previously discussed, priests do not have a significant effect on the probability
of attending a non-Catholic vis-a-vis a Catholic school.
Using these probit models, I include the inverse of the Mills ratio in the second stage
regression (columns (9) and (10) in Table 8). From these regressions, I obtain estimates
that allow me to evaluate whether the effect of an additional voucher school is greater
on public schools or existing voucher schools. The results show that the effect of an
additional voucher school is quantitatively similar for public and voucher schools.
The results in the last two columns of Table 8 confirm the prediction of Corollary 2 in
my model: public schools may react similarly to the entry of an additional voucher school
28
if most public schools face competition from at most one voucher school as the data for
Chile confirm and the difference between the minimum quality offered by public schools
and the minimum quality offered by voucher schools is sufficiently high. As previously
discussed, an alternative interpretation of these results is that either voucher schools do
not tend to cream-skim, peer-effects do not have a causal impact on test scores, or the
incentive effects created by the voucher system on public school students dominate any
cream-skimming effect.
Finally, I present quantile regression estimates of voucher-school competition for
students in different positions of the distribution. Results suggest that the estimated
effects do not vary a lot across quantiles, but are slight smaller for the students in the
1st and 10th quantile than for the other students. Therefore, these results also suggest
that the effect of voucher-school competition on test scores do not vary significantly
accordingly to the students characteristics.
Overall, the evidence presented in this section presents a consistent pattern of posi-
tive effects of voucher-school competition on test scores for the average student and for
students attending public and voucher schools, as predicted by the model. In the next
subsection, I study the effects of voucher-school competition on expenditures, produc-
tivity, and student composition at the school level.
7.2 Expenditures, Productivity, and Student Composition
This subsection studies the role that expenditures in education play in my previous
results. I present results for the effects of voucher school competition on private, public,
municipal (i.e. not related to the voucher or other central government programs), and
total expenditure per capita in columns (1) to (4) of Table 10. The main motivation
for these exercises is to study whether competition increases test scores by increasing
expenditure in education. Results imply positive and significant effects on municipal,
total public, and total expenditures per student. Only the implicit elasticity of the
increase in municipal transfers is economically relevant (1.04), in all the other cases the
implicit elasticities are smaller than 0.20. This result is interesting because suggests
that local governments are able to increase expenditure in education, as competitive
pressures increase. This has the positive effect that the increase in expenditure may
reflect a higher effort to increase quality as a response to political pressures, but, at the
same time, as suggested in my theoretical discussion, non-voucher transfers may dampen
the incentive effect created by competition on test scores. I study the latter in the next
section using a potentially valid source of exogenous variation of non-voucher transfers
29
to local governments.
Next, column (5) in Table 10 study how competition affects the copayment level in
voucher schools. In the model I assume that there are no prices because copayments
represent in average less than one third of public expenditure in education. In this
sense, the model can be interpreted in terms of the effort that agents exert. However,
in a more general model, voucher schools subject to competition may decrease prices.
This is exactly what results in column (5) show: an increase in competition decreases
copayment charged. Interestingly, the implicit elasticity is also economically significant:
a one-percent increase in competition decreases copayments by 0.3%.
Overall, results in column (3) imply that an increase in competition increases total
expenditure per-capita. However, the increase is relatively small: a one-standard devia-
tion increase in competition increases total expenditure per-capita by about 9%. Given
these results, in columns (6) and (7) I study the effects of competition on the produc-
tivity of education expenditure. I define two proxies for productivity: the (normalized)
ratio of test scores to total expenditure per student and the log of total expenditure
per student, respectively. These are rough measures of productivity–i.e. test scores
per unit of expenditure—that can be interpreted as proxies for agents effort, their key
choice variable in my theoretical model. Results suggest that the effects of voucher-
school competition on both measures of productivity are positive and significant. An
increase of one-standard deviation of voucher school competition increases normalized
0.20 and 0.23 standard deviations of the productivity indexes. These results imply that
voucher-school competition increases the productivity of schools, as emphasized in the
theoretical model of this paper.
The last three columns of Table 10 study how competition affects the composition of
students within schools. Again, this is not present in my theoretical model, but exten-
sions where parents and/or school agents value homogeneity in their schools may predict
that increases in competition decrease the dispersion of student characteristics within
schools.27 I study this idea using an index of the distance of each student to the average
student in her school. I use two measures of socioeconomic characteristics (mother’s
education and per-capita income) and test scores as measures of student characteris-
tics My results are not conclusive about the effect of competition on test scores, but
imply a negative and significant effect of competition on the heterogeneity of mother’s
27Results in Gallego and Hernando (2006) support this presumption: in equilibrium students inSantiago tend to attend schools with similar peers in terms of mother’s education, income, gender, andpreferences for the teaching of values. Work in progress is trying to disentangle the effects of supplyand demand factors to explain this equilibrium allocation of students.
30
education within schools, a negative but insignificant effect on heterogeneity of income
within schools, and a positive and marginally significant effect on the heterogeneity of
test scores within schools.
8 Estimating Interaction Effects
In this section, I expand the previous analysis by studying the implication of my model
that there should be interaction effects: public school response to exogenous changes
in voucher school competition depends on how binding the minimum enrollment is.
Corollary 3 predicts that the effect of voucher school entry on test scores is smaller
for schools that have low minimum enrollment levels (which in the model corresponds
to receiving big non-voucher transfers). If this is the case, public schools can meet
the minimum enrollment constraint more easily. This result illustrates the theoretical
mechanism at work in the model, which is related to the implicit incentives that the
minimum enrollment constraint presents for public school agents. In this section, I test
this prediction against the data using proxies for (i) the size of the education deficit
as a percentage of education revenues (a proxy for non-voucher transfers) and (ii) the
average size of public schools in different municipalities. I interpret these characteristics
as the degree of softness of the public school budget constraints.28 Using these proxies,
I study whether differences in these proxies affect the response of public schools to
voucher-school competition. To do so, I split the sample of public schools into those
located in municipalities that have education deficits above and below the median, and
public schools located in municipalities that have average school sizes above and below
the median of the distribution. I expect the effect of voucher-school competition to be
larger in the samples in which education deficits are relatively low and average school
size are relatively big.
I use a short-lived change in the Chilean electoral law that allows me to identify
the short-run variation in deficits and average school size and, therefore, to control for
potential selection bias in my estimates. The electoral law operating between 1999 and
2003 establishes that a mayor (who manages the public schools) is the elected member
of the municipal council who receives the most votes (conditional on getting at least 20%
of the votes). This source of variation is useful for my identification strategy because,
28In general, non-voucher transfers are not allocated to local government in a transparent way. Serranoand Berner (2002) document the transfer process for the case of local government education debts relatedto teacher pensions. As the authors document, it is not easy to follow the actual decision process inpart because central government authorities did not want to establish precedents to be used by othermayors.
31
as previously discussed, municipalities receive discretionary transfers from the central
government and pro-government mayors are able to raise more of these funds.
I implement a difference-in-difference regression to study the effects of the 1999 elec-
toral law on my two proxies for the degree of bindingness of the minimum enrollment
level in the context of a selection model of the form:
Pm = 1³ϕVmKm + Vm + φKm + ζZm +M
0mκ + Y /
mτ + µm > 0´, (6)
where Pm is an indicator function that takes a value of one if the municipality has an
education deficit above the median or an average public school size below the median;
Vm is the share of votes that goes to the pro-government coalition; and Km is a dummy
that takes a value of one if the mayor belongs to the pro-government coalition. I exclude
the interaction of Vm and Km from (4) and include each variable separately, as well as
the estimated inverse of the Mills-ratio in (4). Table 11 presents my marginal probit
estimates of equation (6). The results indicate that the interaction variable has a positive
and significant effect: the probability that a municipality has an education deficit above
the median increases by 50% and average school size below the median increases by
about 40% if the mayor belongs to the pro-government coalition and the probability of
having, given the vote obtained by the pro-government coalition.
Table 12 presents estimated interaction effects. I present estimates including selection
correction in the top panel and without including selection correction in the bottom
panel. Results in both cases are qualitatively similar. The results for the subsample
of students attending public schools that have high education deficits and low average
school sizes tend to react by less than the other schools (and only in one case the
reaction is positive and statistically significant). Overall, these results show that proxies
for the bindingness of the minimum enrollment level affect the degree of response of
public schools to voucher-school competition, as predicted by my model and support
the existence of heterogeneous effects of voucher school competition on public schools.
These results are hard to reconcile with alternative explanations for the positive effects
of voucher school competition.
9 Concluding Comments
The potential effects of school vouchers and inter-school competition on student out-
comes has been a much debated topic in the US and elsewhere. My study of the Chilean
voucher system, which has operated for more than 20 years in the complete K-12 system,
can help us to understand the effects of vouchers on educational outcomes. Previous
32
research has been stymied by endogeneity problems. I argue that the interaction of
the variation in the number of priests per person across Chilean areas in 1950, and the
institution of the voucher system in 1981, allows me to identify the effects of voucher-
school competition on test scores. I document that the number of Catholic priests is
not correlated with educational outcomes in the pre-voucher period and is correlated
with educational outcomes in the post-1981 period. This result allows me to use the
number of priests per person in 1950 as an instrument for voucher-school entry during
the voucher period.
I find that once I instrument for the ratio of voucher-to-public schools in an area, one
additional voucher school per public school increases test scores by about 0.14 standard
deviations. The magnitude of this effect on test scores is equivalent to about half of the
effect of increasing a mother’s attainment from primary to secondary education. These
results are roughly similar for students attending public schools and students attending
non-Catholic voucher schools.
My estimates of the effects of school competition on test scores are smaller for stu-
dents attending public schools that face less binding minimum enrollment, measure using
two alternative proxies. While agents operating voucher schools receive higher payoffs if
they increase enrollment, agents operating public schools receive fixed wages and only
have to meet a minimum enrollment constraint. Therefore, agents operating in areas
where the minimum enrollment constraint is less binding react less to voucher-school
competition, as predicted by the theoretical model presented in the paper. Overall, the
evidence is consistent with a theoretical rationale that emphasizes the role of incentives
provided by voucher-school competition.
My results do not imply that selection or segregation are not relevant issues in the
Chilean case. Rather, controlling for characteristics of students and markets, there are
sizeable direct effects of competition on test scores. More than 20% of educational mar-
kets in Chile have no voucher school in operation. Similarly, there are heterogeneous
effects of voucher school competition for public school students, depending on how bind-
ing minimum enrollment constraints are. Thus, the introduction of the voucher system
does increase educational inequality in Chile. The paradox, though, is that the Chilean
system does not become more unequal because of the existence of voucher schools, but
rather because of the absence of voucher schools in some areas, and the absence of strong
incentives for some public school agents. The government could correct this inequality
while preserving school choice by using the right incentives, such as letting per-student
subsidies depend upon student characteristics, as proposed by Gonzalez et al. (2002)
33
Hoxby (2001), and Sapelli (2003), or by creating explicit incentives that relate the welfare
of public school agents to student outcomes.
A Appendix: Proofs
A.1 Result 1
First, notice that if N = 0, nP = L.
Second, I analyze the case when N = 0. Define x as the location of parents that
are indifferent toward both types of schools. These parents determine the share of the
market going to each school. x is given by:
x =qP − qV + t
2t=
qP − qV2t
+1
2.
This expression implies that the number of students attending public and voucher schools
are xL and (1− x)L, respectively. Therefore,
nP = LqP − qV2t
+L
2, and
nV = LqV− q
P
2t+
L
2.
Regarding the distribution of students among voucher schools. If N = 1 obviously the
enrollment of the only market school in the market is nV . If N ≥ 1, since parents
randomize among all schools that offer the same quality, the probability that each of
them is selected is 1N, which implies that:
nV i =nVN
if N ≥ 1.
A.2 Result 2
Since quality is costly and that the only incentive to increase quality above the minimum
level is to meet the minimum enrollment constraint, always one of the two constraints
will bind in equilibrium. The simplest case is when N = 0, then optimal public school
quality is qMP .
If N > 0 and the minimum quality constraint binds (i.e. Condition 1 does not hold),
qP = qMP . This will be the case if:
(n
L− 12) <
qMP − qV2t
. (7)
34
In this case, the number of students attending public schools is:
nP = LqMP − qV2t
+L
2.
If Condition 1 holds, then the minimum enrollment constraint binds and, therefore,
nP = n. Given Result 1, optimal public school quality is:
qP = 2t
µn
L− 12
¶+ qV .
A.3 Result 3
Voucher schools enter the market if:h(v − cV (qV ))
nVN− F
i≥ 0.
N∗ = 0 if
(v − cV¡qMV¢)nV < F.
because in this case, even producing the lowest possible level produces negative profits.
Using a similar argument, N∗ = 1 when:
(v − cV¡qMV¢)nV ≥ F , and
(v − cV¡qMV¢)nV2
< F.
Analogously, N∗ = 2 when:
(v − cV¡qMV¢)nV2≥ F , and
(v − cV¡qMV¢)nV3
< F.
Generalizing this argument, N∗ has to satisfy:
(v − cV¡qMV¢)nVN≥ F , and
(v − cV¡qMV¢)
nVN + 1
< F.
Rearranging terms:
(v − cV¡qMV¢)
nVN + 1
< F ≤ (v − cV¡qMV¢)nVN
.
35
A.4 Result 5
If Condition 1 holds, N = 1, and NP = 1, voucher schools face no competition from
other voucher schools and public schools will respond to their choices in order to meet
the minimum enrollment constraint. Therefore, voucher schools face a constant demand
nV = (L− n), for any level of quality they offer. Then, profit maximization implies that
voucher schools offer:
qV = qMV .
If N = 1, but NP = 2, and Condition 1 holds, the incumbent voucher has to offer a
quality level such that the potential entrant is indifferent, i.e., profits have to be 0. Thus,
qV = cV−1µv − NF
L− n
¶.
If Condition 1 does not hold, and N = 1, NP = 1, the only difference with the previous
case is that enrollment in the voucher school is not constant. Moreover, Result 1 implies
that voucher school enrollment is increasing in voucher school quality (notice that the
public school produces qMP . Therefore, the optimization of the voucher school is:
MaxqV
∙(v − cV (q
V))L
µqV− qM
P
2t+1
2
¶− F
¸,
the first order condition of this problem is (assuming an interior solution):
cV0(qV )nV −
cV (qV )L
2t= 0⇔ cV (qV )
cV 0 (qV )= 2t
nV
L.
Finally, If Condition 1 does not hold, and N ≥ 1, and NP ≥ 2, competition betweenvoucher schools imply that profits of incumbents go to 0, and, therefore:
qV = cV−1µv − NF
nV
¶.
References
[1] Aedo, C. (1998). “Diferencias entre escuelas y rendimiento estudiantil en Chile”,
cap. 2 in La Organizacion Marca la Diferencia: Educacion y Salud en America
Latina”, Research Network, IADB.
[2] Aedo-Richmond, C. (2000) La educacion privada en Chile : un estudio historico-
analıtico desde el perıodo colonial hasta 1990. Santiago, Chile: RiL Editores.
[3] Aliaga, F. (1989). La Iglesia en Chile : Contexto Historico. Santiago, Chile: Edi-
ciones Paulinas.
36
[4] Araneda, F. (1986). Historia de la Iglesia en Chile. Santiago, Chile: Ediciones
Paulinas.
[5] Auguste, S. and J.P. Valenzuela (2003), ”Do students benefit from school competi-
tion? Evidence from Chile”, Mimeo, University of Michigan.
[6] Barrios, M. (1992). Chile y su Iglesia : una sola Historia. Santiago, Chile: Editorial
Salesiana.
[7] Barro, R. and R. McCleary (2003). “Religion and Economic Growth”. NBERWork-
ing Paper 9682.
[8] Bayer, P. and R. McMillan (2005). ”Choice and Competition In Local Education
Markets”. Mimeo, Yale University and University of Toronto, July.
[9] Bellei, C., G. Munoz, L. Perez and D. Raczynski (2003) Escuelas Efectivas en
Sectores de Pobreza:¿Quien dijo que no se puede? UNICEF and Government of
Chile.
[10] Carnoy, M. and P. McEwan (1998). ”Is private education more effective and cost-
effective than public? The case of Chile”. Mimeo, Universidad de Stanford.
[11] CENDA (2002). ”Estudio de Remuneraciones del Magisterio” Manuscript, CENDA,
December.
[12] Contreras, D. and V. Macias (2002). ”Competencia y resultados educacionales”.
Mimeo, University of Chile.
[13] Checchi, D. and T. Jappelli (2004) ”School Choice and Quality”, Working Paper
91. Universita Degli Studi di Salerno.
[14] Espinola, V. (1993). ”The educational reform of the military regime in Chile: The
school system’s response to competition, choice, and market relations. Ph.D. dis-
sertation, University of Wales College Cardiff.
[15] Gallego, F. (2002). “Competencia y Resultados Educativos: Teorıa y Evidencia
para Chile” Cuadernos de Economıa, 39 (118): 309-352.
[16] Gallego, F. and A. Hernando (2006). ”School Choice in Chile: Looking at the
Demand Side”, Work in Progress, Harvard and MIT.
37
[17] Garrone (1977). “The Catholic School”. Manuscript, The Sacred Congregation for
Catholic Education, Rome.
[18] Gonzalez, P., Mizala, A. and P. Romaguera (2002). “Recursos diferenciados a la
educacion subvencionada en Chile”. Working Paper # 150, Centro de Economıa
Aplicada, University of Chile.
[19] Grossman, M. (2005). ”Eduaction and Nonmarket Outcomes”. NBER Working Pa-
per 11582.
[20] Gruber, J. (2005). ”Religious Market Structure, Religious Participation, and Out-
comes: Is Religion Good for You?” NBER Working Paper 11377.
[21] Hanushek, E. A., and Kimko, D. (2000). Schooling, labor force quality, and the
growth of nations. American Economic Review, 90(5): 1184-1208.
[22] Hotelling, (1929). ”Stability in Competition”, The Economic Journal, 39 (153):
41-57.
[23] Hoxby, C. (1994). “”Do Private Schools Provide Competition for Public Schools?”
NBER Working Paper 4978.
[24] Hoxby, C. (2000). “Does Competition Among Public Schools Benefit Students and
Taxpayers?” American Economic Review 90 (5): 1209-1238.
[25] Hoxby, C. (2001) “Ideal Vouchers”. Manuscript, Harvard University.
[26] Hoxby, C. (2005) ”Competition Among Public Schools: A Reply to Rothstein
(2004)”, NBER Working Paper 11216.
[27] Howell, W. G. (2004), “Dynamic selection effects in a means-tested, urban school
voucher program,” Journal of Policy Analysis and Management 22(3), 225-50.
[28] Hsieh, C. and M. Urquiola (2004). “When school compete, how do they com-
pete? An assesment of Chile’s nationwide school voucher program. Manuscript,
UC-Berkeley University and Columbia University.
[29] Juerges, H., W. Richter, y K. Schneider (2004) “Teacher Quality and Incentives:
Theoretical And Empirical Effects Of Standards On Teacher Quality” Mimeo, CE-
Sifo.
38
[30] Lavy, V. (2005). ”From Forced Busing to Free Choice in Public Schools: Individual
and General Equilibrium Effects”. Manuscrispt, Hebrew University.
[31] Lehmann, C. and X. Hinzpeter (1997). ”Opinion Publica y Educacion en Chile:
Preferencias y Disposiciones”, Puntos de Referencia 183, Centro de Estudios Pub-
licos.
[32] MINEDUC (1975). ”Informe de Matricula de la Ensenanza Pre-Basica, Especial y
Basica Ninos” Mimeo, Ministerio de Educacion, Chile.
[33] Mizala, A. and P. Romaguera (2004). “Teachers’ Salary Structure and Incentives in
Chile”. Working Paper # 193, Centro de Economıa Aplicada, University of Chile.
[34] Pasalacqua, A. (2004). ”Estadisticas Educacion Catolica 2002”, Conferencia Epis-
copal de Chile.
[35] Robinson, J. and R. Torvik (2005). ”A Political Economy Theory of the Soft Budget
Constraint”, CEPR Discussion Paper No. 5274
[36] Rothstein, J. (2004). ”Good Principals or Good Peers: Parental Valuation of School
Characteristics, Tiebout Equilibrium, and the Incentive Effects of Competition
Among Jurisdictions.” NBER Working Paper #10666.
[37] Rothstein, J. (2005). ”Does Competition Among Public Schools Benefit Students
and Taxpayers? A Comment on Hoxby (2000)” NBER Working Paper 11215.
[38] Sander, W. (2001), “The effect of Catholic schools on religiosity, education and com-
petition,” National Center for the Study of Privatization in Education Occasional
Paper #32.
[39] Sandstrom, F. and F. Bergstrom (2005) “School Vouchers in Practice: Competition
will not Hurt You”, Journal of Public Economics 89: 351-380.
[40] Sapelli, C. (2003). “The Chilean Voucher System: Some New Results and Research
Challenges”. Cuadernos de Economıa, 40 (121): 530-538.
[41] Sapelli, C. and B. Vial (2002). “The Performance of Private and Public Schools in
the Chilean Voucher System”. Cuadernos de Economıa, 39 (118): 423-454.
[42] Sapelli, C. and B. Vial (2003). “Peer Effects and Relative Performance of Vouchers
Schools in Chile”. Working Paper # 256, Economics Department, Catholic Univer-
sity of Chile.
39
[43] Serrano, C. and H. Berner (2002). ”Chile: Un Caso Poco Frecuente de Indisciplina
Fiscal (Bailout) y Endeudamiento Encubierto de la Educacion Municipal”, Working
Paper R-446, IADB.
[44] Soto, R. and A. Torche (2004). ”Spatial Inequality, Migration and Economic Growth
in Chile”. Cuadernos de Economıa, 41: 401-424.
[45] Tirole, J. (1988). The Theory of Industrial Organization. The MIT Press.
[46] Tironi, M. (2003). Nueva Pobreza Urbana: Vivienda y Capital Social en Santiago
de Chile, 1985-2001. Ril Editores.
[47] Wooldridge, J. Econometric Analysis of Cross Section and Panel Data. The MIT
Press.
40
Variable Obs. Mean Median Standard Deviation
5th percentile
95th percentile
Source
Test scores (standardized) 187610 -0.03 -0.03 0.97 -1.71 1.49 SIMCEMother's education
Primary 187610 0.35 0.00 0.48 0.00 1.00 SIMCESecondary-General 187610 0.32 0.00 0.47 0.00 1.00 SIMCE
Secondary-Technical 187610 0.16 0.00 0.37 0.00 1.00 SIMCEHigher-General 187610 0.08 0.00 0.27 0.00 1.00 SIMCE
Higher-Technical 187610 0.09 0.00 0.28 0.00 1.00 SIMCEAverage Years of Schooling 187610 9.84 11.00 3.31 5.00 15.00 SIMCE
Log(Per-Capita Income) 187610 10.27 10.31 0.90 8.87 11.83 SIMCEValues among the three most important reasons for choosing a school 187610 0.28 0.00 0.45 0.00 1.00 SIMCEVoucher school 187610 0.43 0.00 0.50 0.00 1.00 SIMCECatholic voucher school 187610 0.15 0.00 0.39 0.00 1.00 http://www.iglesia.cl/Public expenditure per-capita 161563 27451 22128 12986 19219 52540 SIMCE, http://www.sinim.cl/, MINEDUCPrivate expenditure per-capita 180288 19309 12500 21548 2500 60000 SIMCEMunicipal transfers per-capita 96564 4115 1688 8332 0 12698 http://www.sinim.cl/Total expenditure per-capita 155197 46643 40971 22261 23465 87582 Author's calculationsCopayments 183035 5730 2500 10288 0 25000 SIMCEProductivity index 1 (test scores divided by total expenditure) 155197 -0.04 0.03 0.99 -1.79 1.47 Author's calculationsProductivity index 2 (test scores divided by log total expenditure) 155197 0.02 0.08 1.00 -1.71 1.58 Author's calculations
High school graduation dummy 75805 0.40 0.00 0.49 0.00 1.00 CASENImmigration dummy 8857 0.17 0.00 0.37 0.00 1.00 Social Protection Survey
Ratio of Voucher to Public Schools 298 0.47 0.25 0.70 0.00 1.6 Ministry of EducationMean years of schooling 298 8.82 8.66 1.29 7.21 10.96 SIMCEStandard Deviation of Years of schooling 298 3.45 3.47 0.47 2.70 12 SIMCEMean Log Income 298 10.02 9.94 0.37 9.55 10.71 SIMCEStandard Deviation of Log Income 298 0.85 0.85 0.14 0.66 1.08 SIMCELog School-Age Population 298 8.38 8.36 1.28 6.32 10.7 http://www.sinim.cl/Urbanization Rate 298 0.34 0.25 0.28 0.00 0.93 Ministry of EducationHigh-School Graduation Rate
37<age 304 0.28 0.26 0.14 0.08 0.56 CASEN25<age<38 304 0.46 0.45 0.17 0.20 0.74 CASEN
age<26 304 0.52 0.53 0.16 0.24 0.79 CASENPro-Government Mayor 331 0.51 0.00 0.50 0.00 1.00 Electoral OfficePro-Government Vote 331 0.53 0.52 0.15 0.25 0.78 Electoral Office
Priests per 1000 people in 1950 26 0.32 0.31 0.13 0.15 0.54 Annuario PontificioPriests per 1000 people in 2000 26 0.15 0.14 0.04 0.08 0.21 Annuario PontificioCatholic Affiliation 26 0.78 0.79 0.08 0.68 0.90 Annuario PontificioRatio of Religious to Total Priests 26 0.45 0.44 0.16 0.17 0.66 Annuario Pontificio
High-School Graduation Rate39<age 26 0.27 0.27 0.17 0.05 0.5 Social Protection Survey
27<age<40 26 0.47 0.59 0.22 0.17 0.71 Social Protection Surveyage<28 26 0.60 0.65 0.24 0.22 1 Social Protection Survey
High-School Graduation Rate in Public Schools39<age 26 0.26 0.27 0.16 0.08 0.5 Social Protection Survey
27<age<40 26 0.45 0.54 0.20 0.17 0.67 Social Protection Surveyage<28 26 0.55 0.62 0.26 0.00 1 Social Protection Survey
Voucher school attendance (% of total population)39<age 26 0.05 0.05 0.04 0.00 0.12 Social Protection Survey
27<age<40 26 0.08 0.07 0.06 0.00 0.23 Social Protection Surveyage<28 26 0.11 0.12 0.09 0.00 0.27 Social Protection Survey
Notes: Detailed definitions of each variable appear in the main text
Region x Urban/Rural variables
Table 1: Descriptive Statistics
Student-level variables
Market-level variables
Diocese-level variables
Individual-level variables
Dependent Variable:
(1) (2) (3) (4)Log(income) 0.28 0.10
(0.27) (0.26)Schooling 0.13 0.06
(0.09) (0.09)1.25 1.19
(0.52) (0.53)R2 0.0462 0.0783 0.2415 0.2494Number of dioceses 26 26 26 26Cross section regressions, each observation represents the value for a dioceses. Robust standard errors in parentheses. Constants are not reported
Ratio of order to total priests
Table 2Determinants of Priests per Capita in 1950:
Religious Order EffectsLog of Priests per 1,000 people in 1950
Dependent Variable:
Sample: Age >37 37>=Age>=26 Age<26
(1) (2) (3)0.05 0.08 0.17
(0.13) (0.14) (0.10)Catholic affiliation 0.02 0.10 0.01
(0.07) (0.35) (0.25)Urbanization Rate 0.29 0.33 0.26
(0.03) (0.04) (0.03)-0.02 0.01 0.040.05 (0.03) (0.02)
R2 0.49 0.55 0.46Number of markets 300 300 300Cross section regressions, each observation represents a value a for a market. Standard errors clustered at the diocese level in parenthesis. Region dummies and constants are not reported.
Log(population)
Log(Priests per 1000 people)
Table 3Municipal level regressions for
eduactional outcomes before and after the reform
High-School Graduation Rate
Dependent Variable: Share of enrollment in
voucher schools
Change in Public School
Enrollment (Post-Pre Reform)
Ratio of Catholic to
public schools
Ratio of Non-catholic to public
schools
Share of Non-Catholic Schools (among Voucher
Schools)
(1) (2) (3) (4) (5) (6) (7)Mean of mother’s schooling 0.22 0.21 0.04 -0.03 0.05 0.15 -0.03
(0.08) (0.09) (0.01) (0.01) (0.02) (0.07) (0.05)-0.16 -0.07 0.02 -0.09 -0.06 -0.17 0.09(0.23) (0.33) (0.04) (0.04) (0.04) (0.23) (0.09)
Mean of Log Income 0.16 0.20 -0.02 0.00 0.00 0.17 0.02(0.13) (0.11) (0.02) (0.01) (0.01) (0.12) (0.03)-1.20 -1.28 -0.11 0.07 -0.15 -1.27 -0.13(0.24) (0.55) (0.07) (0.08) (0.09) (0.35) (0.14)
Log(Population) 0.22 0.21 0.05 -0.03 0.02 0.19 0.02(0.08) (0.09) (0.01) (0.01) (0.02) (0.07) (0.02)
Urbanization Rate 0.38 0.30 0.17 -0.16 0.02 0.00 0.13(0.23) (0.23) (0.05) (0.06) (0.01) (0.03) (0.12)0.50 0.06 -0.08 0.05 0.36 0.03
(0.10) (0.02) (0.02) (0.02) (0.08) (0.02)1.69
(0.69)1.69 2.99 0.87 -0.71 0.90 0.54 -0.55
(0.69) (0.75) (0.14) (0.14) (0.27) (0.27) (0.20)R2 0.72 0.76 0.71 0.69 0.47 0.61 0.25Number of markets 298 298 298 298 298 298 203
Availability of Voucher Schools
Share of Catholics in total population
Cross section regressions using the actual first stages. Clustered standard errors at the diocese level in parenthesis. Only estimates for variables at the market level are reported.
Ratio of order to total priests
Ratio of voucher to public schools
Standard Deviation of mother's schooling
Standard Deviation of Log Income
Log(Priests per 1,000 people in 1950)
Table 4
Dependent Variable:
Estimation Technique
Urban area*post
Pseudo-R2/R2
Number of observationsCross section regressions weighted by expansion factors for individuals aged 19-50. Clustered standard errors at the diocese level in parenthesis. Estimates for age dummies, region dummies, and the interaction for region dummies and the post reform dummy are not reported.
0.0810566
0.0610566
Probit LPM
(0.03)
0.26(0.12)
0.00(0.01)
0.07(0.01)
-0.02
0.10(0.04)0.29
Log(Priests per 1,000 people in 1950) -0.04(0.12)
(0.07)0.09Urban area
(0.02)
Log(Priests per 1,000 people in 1950)*Post Reform
Table 5Attendance to "Voucher" Schools, Pre and Post
ReformDummy=1 if Person attended a subsidized private
school
(1) (2)
Dependent Variable: (1) (2) (3) (4)
0.04 0.03(0.10) (0.13)
0.20 0.09(0.24) (0.31)
Econometric TechniqueNumber of observations
Table 6Priests and In-Migration Decisions
In-migration Dummy In-migration Rates
Marginal Probit 8857
OLS Estimates13
Standard errors clustered at the region level in parenthesis. Constants are not reported.
Log(Priests per 1000 people, destination/origin) Ratio order to total priests, destination-origin
Dependent Variable: (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Estimation Technique OLS OLS IV IV IV IV IV IV IV IVMeasure of Competition Ratio of
Voucher to Public
Schools
Ratio of Voucher to
Public Schools
Ratio of Voucher to
Public Schools
Ratio of Voucher to
Public Schools
Ratio of Voucher to
Public Schools
Ratio of Voucher to
Public Schools
Ratio of Voucher to
Public Schools
Enrollment in Voucher Schools
Herfindahl index
Change in Public School
Enrollment (Post-Pre Reform)
Voucher school competition 0.04 -0.02 0.17 0.15 0.16 0.13 0.14 0.98 12.17 -0.93(0.01) (0.02) (0.06) (0.05) (0.04) (0.04) (0.03) (0.37) (4.44) (0.36)
Mother Education:Secondary-General 0.27 0.28 0.28 0.28 0.27 0.27 0.28
(0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01)Secondary-Technical 0.40 0.40 0.40 0.40 0.40 0.40 0.40
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02)Higher-General 0.64 0.64 0.64 0.64 0.64 0.61 0.64
(0.02) (0.02) (0.02) (0.02) (0.02) (0.03) (0.02)Higher-Technical 0.49 0.50 0.50 0.50 0.50 0.47 0.50
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02)Log(Per-Capita Income) 0.23 0.23 0.23 0.23 0.24 0.24 0.23
(0.00) (0.00) (0.00) (0.00) (0.01) (0.00) (0.00)Market Level Variables:
Mean Education 0.02 -0.01 -0.01 -0.01 -0.03 0.09 -0.01(0.02) (0.02) (0.03) (0.02) (0.03) (0.03) (0.02)0.03 0.05 0.05 0.05 -0.03 -0.16 -0.06
(0.04) (0.07) (0.07) (0.07) (0.06) (0.12) (0.06)0.18 0.12 0.13 0.13 0.14 0.17 0.14
(0.03) (0.07) (0.05) (0.06) (0.05) (0.08) (0.05)-0.27 -0.02 -0.05 -0.04 -0.12 -0.25 -0.14(0.05) (0.11) (0.08) (0.08) (0.07) (0.23) (0.06)
Log(Population) -0.03 -0.06 -0.06 -0.06 -0.06 -0.38 -0.06(0.02) (0.02) (0.02) (0.02) (0.01) (0.09) (0.01)
Urbanization Rate 0.00 -0.08 -0.07 -0.07 -0.20 0.09 -0.18(0.09) (0.14) (0.12) (0.13) (0.17) (0.18) (0.17)
Share of Catholic Population -0.20 -0.48 -0.44 -0.46 -1.00 -1.32 -0.89(0.17) (0.26) (0.21) (0.22) (0.41) (0.56) (0.35)
Instrumental Variable: - - Priests Priests Order Order Priests, Order Priests Priests Priests
Over-identification test (p-value) - - - - - - 0.83 -Number of students 187610 187610 187610 187610 187610 187610 187610 187610 187610 187610Number of schools 5433 5433 5433 5433 5433 5433 5433 5433 5433 5433Number of markets 298 298 298 298 298 298 298 298 298 298
Test Scores: OLS and IV ResultsTable 7
Cross section regressions. Standard errors clustered at the diocese level in parenthesis. Region dummies and constants are not reported.
Standard Deviation of Education
Standard Deviation of Log(Per-Capita Income)
Mean Log(Per-Capita Income)
Test scores
Dependent Variable: (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Voucher school competition 0.17 0.21 0.08 0.14 0.17 0.12 0.13 0.13 0.15 0.13(0.05) (0.05) (0.02) (0.04) (0.06) (0.04) (0.04) (0.05) (0.04) (0.05)
Mother Education:Secondary-General 0.27 0.27 0.27 0.27 0.26 0.26 0.26 0.18 0.23 0.24
(0.02) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.02)Secondary-Technical 0.40 0.40 0.41 0.40 0.39 0.40 0.40 0.28 0.34 0.36
(0.02) (0.02) (0.02) (0.01) (0.02) (0.02) (0.02) (0.02) (0.02) (0.03)Higher-General 0.62 0.63 0.64 0.65 0.63 0.64 0.64 0.40 0.53 0.57
(0.02) (0.02) (0.02) (0.01) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02)Higher-Technical 0.48 0.49 0.51 0.50 0.48 0.49 0.49 0.30 0.40 0.43
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.03)Log(Per-Capita Income) 0.24 0.24 0.24 0.24 0.23 0.23 0.23 0.13 0.16 0.20
(0.00) (0.00) (0.01) (0.00) (0.00) (0.00) (0.00) (0.00) (0.01) (0.01)
School Level Variables:Mean Education 0.10
(0.01)-0.01(0.01)0.19
(0.02)-0.14(0.04)
Market Level Variables:Mean Education 0.07 -0.06 0.00 -0.03 -0.03 -0.01 -0.03 -0.06 -0.04 0.00
(0.03) (0.03) (0.03) (0.02) (0.03) (0.02) (0.01) (0.02) (0.03) (0.05)-0.01 0.09 0.06 0.07 0.03 0.05 0.01 0.08 0.04 0.10(0.02) (0.09) (0.06) (0.05) (0.06) (0.07) (0.07) (0.07) (0.05) (0.13)-0.13 0.12 -0.04 0.08 0.13 0.06 0.12 -0.08 0.23 0.11(0.10) (0.07) (0.12) (0.04) (0.06) (0.07) (0.06) (0.04) (0.06) (0.09)0.12 0.06 0.11 0.13 0.03 -0.06 -0.03 0.18 -0.08 -0.22
(0.12) (0.13) (0.15) (0.13) (0.11) (0.10) (0.11) (0.11) (0.05) (0.22)Log(Population) -0.07 -0.06 -0.02 -0.04 -0.05 -0.06 -0.07 -0.06 -0.05 -0.08
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.02) (0.01) (0.01)Urbanization Rate -0.15 -0.13 -0.13 -0.12 -0.18 -0.09 -0.04 -0.05 -0.22 0.20
(0.13) (0.15) (0.11) (0.06) (0.14) (0.13) (0.13) (0.13) (0.12) (0.18)-0.49 -0.62 -0.38 -0.53 -0.54 -0.40 -0.39 -0.49 -0.58 -1.68(0.29) (0.27) (0.21) (0.19) (0.27) (0.26) (0.22) (0.23) (0.23) (0.46)
0.39 0.24(0.06) (0.17)
1.67(0.05)
Instrumental Variable: Priests Priests Priests Priests Priests Priests Priests Priests Priests Priests
Sample
Exclude if parents
attended school after
reform
Exclude if schools located in rural areas
Exclude 90% of Santiago MA
sample, bootstrapping
Exclude 50% of Santiago MA
sample, bootstrapping All students All students All students All students
Include only public school
students, selection model
Include only non-Catholic voucher school students, selection model
Number of students 85819 148860 105045 124517 187610 184784 187545 187610 106094 53319Number of schools 5267 3013 2963 3013 5433 5272 5426 5433 3355 1552Number of markets 298 278 278 298 298 264 292 298 298 176
Share of Catholic Population
SIMCE Tests in 1991
High-School Graduation Rate before the Reform
Test Scores: Additional IV ResultsTable 8
Cross section regressions. Standard errors clustered at the diocese level in parenthesis. Region dummies and constants are not reported.
Standard Deviation of Education
Standard Deviation of Log(Per-Capita Income)
Mean Log(Per-Capita Income)
Test scores
Standard Deviation of Log(Per-Capita Income)
Standard Deviation of Education
Mean Log(Per-Capita Income)
Dependent Variable: Dummy takes a value of 1 if: Student attends a
public school versus a voucher school
Student attends a non-Catholic versus a Catholic voucher
school
(1) (3)-0.36 -0.28(0.01) (0.01)
Mother Education:Secondary-General -0.07 -0.03
(0.01) (0.01)Secondary-Technical -0.10 -0.05
(0.01) (0.02)Higher-General -0.21 -0.07
(0.02) (0.02)Higher-Technical -0.19 -0.06
(0.02) (0.02)Log(Per-Capita Income) -0.11 -0.03
(0.01) (0.01)Market Level Variables:
Mean of Mother Education -0.02 -0.01(0.02) (0.07)-0.09 0.15(0.06) (0.07)-0.11 -0.14(0.01) (0.07)0.00 -0.17
(0.10) (0.25)Log(Population) -0.06 0.09
(0.01) (0.03)Urbanization Rate -0.14 0.08
(0.10) (0.15)-0.07 0.04(0.02) (0.03)
Share of Catholic Population -1.07 -0.63(0.15) (0.22)
Pseudo R2 0.22 0.15Number of students 172309 69937Number of schools 5433 1701Number of markets 285 203
Standard Deviation of Log(Per-Capita Income)
Values among top priorities when choosing among schools
Table 9Choice of Schools: Marginal Probit
Estimates
Ratio of order to total priests
Standard Deviation of Mother Education
Mean of Log(Per-Capita Income)
Log of (Priests per 1,000 people)
Cross section regressions, each observation represents a value a for a student. Standard errors clustered at the diocese level in parenthesis. Region dummies and constants are not reported.
Dependent Variable:
Public Expenditures per student
Private Expenditures per student
Total expenditures per
student
Municipal Expenditures per student Copayment
Normalized Productivity
Index 1
Normalized Productivity
Index 2
Distance of mother's
education to school mean
Distance of per-capita income to
school mean
Distance of test score to school
mean(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Voucher school competition 3761.94 867.71 4464.08 3572.42 -2513.69 0.23 0.20 -0.09 -0.01 0.007(1463.37) (635.96) (1701.04) (493.12) (1067.13) (0.07) (0.07) (0.04) (0.01) (0.003)
Instrumental Variable: Priests Priests Priests Priests Priests Priests Priests Priests Priests Priests
Sample
All students All students All students All students
Voucher school students w/ selection
correction All students All students All students All students All students
Number of students 165163 180288 155197 96564 80148 155197 155197 187538 187610 187610Number of schools 4488 5427 4452 2845 2076 4452 4452 5404 5433 5433Number of markets 261 298 261 222 203 261 261 298 298 176
Expenditures in Education, Productivity, and Student Composition: IV ResultsTable 10
Cross section regressions. Standard errors clustered at the diocese level in parenthesis. Estimates for other covariates are not reported.
Dependent Variable: Dummy takes a value of 1 if
municipality has Education Deficit above
the Medianhas Average School
Size below the Median
(1) (2)0.50 0.41
(0.26) (0.10)-0.07 -0.16(0.14) (0.06)-0.51 -0.30(0.17) (0.07)
Mean of Mother Education -0.14 -0.01(0.03) (0.01)-0.30 0.04(0.08) (0.02)0.25 -0.02
(0.07) (0.02)-0.37 -0.08(0.21) (0.06)
Log(Population) 0.06 -0.06(0.02) (0.01)
Urbanization Rate 0.30 0.13(0.10) (0.03)-0.13 -0.05(0.05) (0.02)
Share of Catholic Population -0.05 -0.22(0.27) (0.09)
Pseudo R2 0.17 0.38Number of municipalities 299 249Cross section regressions, each observation represents a value a for a municipality, estimates from the selection equation actually used in the paper.
Log of (Priests per 1,000 people)
Proxies for Soft Budget Constraints in Public Schools: Marginal Probit Estimates
Standard Deviation of Mother Education
Standard Deviation of Log(Per-Capita Income)
Mean of Log(Per-Capita Income)
Table 11
Pro-Government Mayor
Pro-Government Vote
Pro-Government Mayor*Pro-Government Vote
Sample: Municipalities with
Number of studentsNumber of schoolsNumber of municipalitiesCross section regressions, each observation represents a value a for a student. Standard errors clustered at the diocese level in parenthesis. Coefficients of all other control variables included in the second and first stage equations are not reported.
194
High school size
878602470
Test Scores: IV Estimates, Interaction Effects Table 12
Second stage estimates without selection correction
(0.21)
(4)
0.17(0.03)
(0.20) (0.06)
(2)
0.52
Ratio of voucher to public schools
108
Big education deficits468191309
Ratio of voucher to public schools
(1)
0.08(0.04)
(3)
0.06(0.20)
Second stage estimates with selection correction0.09 0.52 0.03 0.18
(0.07) (0.31)
885104
Low school size
18234
Small education deficits592752046191
top related